CLI Reference#

usage: chemprop [-h] {train,predict,convert,fingerprint,hpopt} ...

mode #

mode: Possible choices: train, predict, convert, fingerprint, hpopt

Path to which the log file should be written (specifying just the flag alone will automatically log to a file chemprop_logs/MODE/TIMESTAMP.log , where ‘MODE’ is the CLI mode chosen, e.g., chemprop_logs/MODE/2026-06-12T18-58-16.log)

-v

Increase verbosity level to DEBUG

Default: False

-q

Decrease verbosity level to WARNING or ERROR if specified twice

Default: 0

--accelerator

Passed directly to the lightning Trainer()

Default: 'auto'

--devices

Passed directly to the lightning Trainer() (must be a single string of comma separated devices, e.g. ‘1, 2’ if specifying multiple devices)

Default: 'auto'

--constraints-path

Path to a CSV file containing the constraints for atomic/bond properties prediction. The file should have one column for each property being constrained with no SMILES column. The order of the rows should match the order of the SMILES in the input CSV. See also –constraints-to-targets for how to specify which constraint applies to which prediction.

--constraints-to-targets

The column names of the atom or bond targets that correspond to each constraint column in the constraints CSV.

--config-path

Path to a configuration file (command line arguments override values in the configuration file)

-i, --data-path

Path to one, two, or three input CSV files containing SMILES and the associated target values. If one data file is supplied, it will undergo train-val-test split; if two are supplied, the first will undergo train-val split and the second will be taken as test data; if three are supplied, they will be taken as train, val, test data respectively

-o, --output-dir, --save-dir

Directory where training outputs will be saved (defaults to CURRENT_DIRECTORY/chemprop_training/STEM_OF_INPUT/TIME_STAMP)

--remove-checkpoints

Remove intermediate checkpoint files after training is complete.

Default: False

--ensemble-size

Number of models in ensemble for each splitting of data

Default: 1

--atom-multiclass-num-classes

Number of classes for atom targets when running multiclass classification

Default: 3

--bond-multiclass-num-classes

Number of classes for bond targets when running multiclass classification

Default: 3

--pytorch-seed

Seed for PyTorch randomness (e.g., random initial weights)

Shared input data args #

-s, --smiles-columns

Column names in the input CSV containing SMILES strings (uses the 0th column by default)

-r, --reaction-columns

Column names in the input CSV containing reaction SMILES in the format REACTANT>AGENT>PRODUCT, where ‘AGENT’ is optional

--no-header-row

Turn off using the first row in the input CSV as column names

Default: False

--ignore-stereo

Ignore stereochemical information (R/S and Cis/Trans) in the input SMILES

Default: False

--reorder-atoms

Reorder atoms in the Chem.Mol object using the specified atom map numbers

Default: False

Dataloader args #

-n, --num-workers

Number of workers for parallel data loading where 0 means sequential (Warning: setting num_workers to a value greater than 0 can cause hangs on Windows and MacOS)

Default: 0

-b, --batch-size

Batch size

Default: 64

Featurization args #

--rxn-mode, --reaction-mode

Possible choices: REAC_PROD, REAC_PROD_BALANCE, REAC_DIFF, REAC_DIFF_BALANCE, PROD_DIFF, PROD_DIFF_BALANCE

Choices for construction of atom and bond features for reactions (case insensitive):

REAC_PROD: concatenates the reactants feature with the products feature
REAC_DIFF: concatenates the reactants feature with the difference in features between reactants and products (Default)
PROD_DIFF: concatenates the products feature with the difference in features between reactants and products
REAC_PROD_BALANCE: concatenates the reactants feature with the products feature, balances imbalanced reactions
REAC_DIFF_BALANCE: concatenates the reactants feature with the difference in features between reactants and products, balances imbalanced reactions
PROD_DIFF_BALANCE: concatenates the products feature with the difference in features between reactants and products, balances imbalanced reactions

Default: REAC_DIFF

--multi-hot-atom-featurizer-mode

Possible choices: V1, V2, ORGANIC, RIGR

Selects the multi-hot atom featurization scheme. This affects both non-reaction and reaction featurization (case insensitive):

V1: Corresponds to the original configuration employed in the Chemprop V1
V2: Tailored for a broad range of molecules, this configuration encompasses all elements in the first four rows of the periodic table, along with iodine. It is the default in Chemprop V2.
ORGANIC: This configuration is designed specifically for use with organic molecules for drug research and development and includes a subset of elements most common in organic chemistry, including H, B, C, N, O, F, Si, P, S, Cl, Br, and I.
RIGR: Modified V2 (default) featurizer using only the resonance-invariant atom and bond features.

Default: V2

--keep-h

Whether hydrogens explicitly specified in input should be kept in the mol graph

Default: False

--add-h

Whether hydrogens should be added to the mol graph

Default: False

--molecule-featurizers, --features-generators

Possible choices: morgan_binary, morgan_count, rdkit_2d, v1_rdkit_2d, v1_rdkit_2d_normalized, charge

Method(s) of generating molecule features to use as extra descriptors

--descriptors-path

Path to extra descriptors to concatenate to learned representation

--descriptors-columns

Column names in the input CSV containing extra datapoint descriptors, like temperature and pressure. See also –descriptors-path.

--no-descriptor-scaling

Turn off extra descriptor scaling

Default: False

--no-atom-feature-scaling

Turn off extra atom feature scaling

Default: False

--no-atom-descriptor-scaling

Turn off extra atom descriptor scaling

Default: False

--no-bond-feature-scaling

Turn off extra bond feature scaling

Default: False

--no-bond-descriptor-scaling

Turn off extra bond descriptor scaling

Default: False

--atom-features-path

If a single path is given, it is assumed to correspond to the 0-th molecule. Alternatively, it can be a two-tuple of molecule index and path to additional atom features to supply before message passing (e.g., --atom-features-path 0 /path/to/features_0.npz) indicates that the features at the given path should be supplied to the 0-th component. To supply additional features for multiple components, repeat this argument on the command line for each component’s respective values (e.g., --atom-features-path 0 path_zero --atom-features-path 1 path_one) or pass each two-tuple in a series (e.g., --atom-features-path 0 path_zero 1 path_one).

--atom-descriptors-path

If a single path is given, it is assumed to correspond to the 0-th molecule. Alternatively, it can be a two-tuple of molecule index and path to additional atom descriptors to supply after message passing (e.g., --atom-descriptors-path 0 /path/to/descriptors_0.npz indicates that the descriptors at the given path should be supplied to the 0-th component. To supply additional descriptors for multiple components, repeat this argument on the command line for each component’s respective values (e.g., --atom-descriptors-path 0 path_zero --atom-descriptors-path 1 path_one) or pass each two-tuple in a series (e.g., --atom-descriptors-path 0 path_zero 1 path_one).

--bond-features-path

If a single path is given, it is assumed to correspond to the 0-th molecule. Alternatively, it can be a two-tuple of molecule index and path to additional bond features to supply before message passing (e.g., --bond-features-path 0 /path/to/features_0.npz indicates that the features at the given path should be supplied to the 0-th component. To supply additional features for multiple components, repeat this argument on the command line for each component’s respective values (e.g., --bond-features-path 0 path_zero --bond-features-path 1 path_one) or pass each two-tuple in a series (e.g., --bond-features-path 0 path_zero 1 path_one).

--bond-descriptors-path

Path to additional bond descriptors to use with the learned bond representations after message passing. The file follows the same format as –atom-descriptors-path, i.e. the file is created using np.savez(‘bond_descriptors.npz’, *E_ds) where E_ds is a list of 2D numpy arrays with shape n_bonds x n_descriptors.

--use-cuikmolmaker-featurization

Use cuik-molmaker package for accelerated atom and bond featurization.

Default: False

transfer learning args #

--checkpoint

Path to checkpoint(s) or model file(s) for loading and overwriting weights. Accepts a single pre-trained model checkpoint (.ckpt), a single model file (.pt), a directory containing such files, or a list of paths and directories. If a directory is provided, it will recursively search for and use all (.pt) files found for prediction.

--freeze-encoder

Freeze the message passing layer from the checkpoint model (specified by --checkpoint).

Default: False

--model-frzn

Path to model checkpoint file to be loaded for overwriting and freezing weights. By default, all MPNN weights are frozen with this option.

--frzn-ffn-layers

Freeze the first n layers of the FFN from the checkpoint model (specified by --checkpoint). The message passing layer should also be frozen with --freeze-encoder.

Default: 0

--from-foundation

Name of pretrained foundation model used to initialize message passing. One of: CHEMELEON, or a path to a local model file.

message passing #

--message-hidden-dim

Hidden dimension of the messages (specify multiple values to customize multicomponent encoders separately)

Default: [300]

--message-bias

Add bias to the message passing layers

Default: False

--depth

Number of message passing steps (specify multiple values to customize multicomponent encoders separately)

Default: [3]

--undirected

Pass messages on undirected bonds/edges (always sum the two relevant bond vectors)

Default: False

--dropout

Dropout probability in message passing/FFN layers

Default: 0.0

--mpn-shared

Whether to use the same message passing neural network for all input molecules (only relevant if number_of_molecules > 1)

Default: False

--aggregation, --agg

Possible choices: mean, sum, norm

Aggregation mode to use during graph predictor

Default: 'norm'

--aggregation-norm

Normalization factor by which to divide summed up atomic features for norm aggregation

Default: 100

--atom-messages

Pass messages on atoms rather than bonds.

Default: False

--activation

Possible choices: RELU, CELU, ELU, GELU, GLU, HARDSHRINK, HARDSIGMOID, HARDSWISH, HARDTANH, LEAKYRELU, LOGSIGMOID, LOGSOFTMAX, MISH, MULTIHEADATTENTION, PRELU, RRELU, RELU6, SILU, SIGMOID, SOFTMAX, SOFTMAX2D, SOFTMIN, SOFTPLUS, SOFTSHRINK, SOFTSIGN, TANH, TANHSHRINK, THRESHOLD

Activation function in message passing/FFN layers.

Default: RELU

--activation-args

Arguments for the activation function (Example: arg1 arg2 key1=value1 key2=value2).

FFN args #

--ffn-hidden-dim

Hidden dimension(s) in the FFN top model. A single value is applied to all layers; multiple values specify per-layer widths (must match –ffn-num-layers)

Default: [300]

--ffn-num-layers

Number of hidden layers in FFN top model (v2 semantics, differs from v1)

Default: 1

extra MPNN args #

--batch-norm

Turn on batch normalization after aggregation

Default: False

--multiclass-num-classes

Number of classes when running multiclass classification

Default: 3

Atom FFN args #

--atom-task-weights

Weights to apply for all atom tasks in the loss function

--atom-ffn-hidden-dim

Hidden dimension(s) in the atom FFN top model

Default: [300]

--atom-ffn-num-layers

Number of layers in atom FFN top model

Default: 1

Bond FFN args #

--bond-task-weights

Weights to apply for all bond tasks in the loss function

--bond-ffn-hidden-dim

Hidden dimension(s) in the bond FFN top model

Default: [300]

--bond-ffn-num-layers

Number of layers in bond FFN top model

Default: 1

Atom constrainer FFN args #

--atom-constrainer-ffn-hidden-dim

Hidden dimension(s) in the atom constrainer FFN top model

Default: [300]

--atom-constrainer-ffn-num-layers

Number of layers in atom constrainer FFN top model

Default: 1

Bond constrainer FFN args #

--bond-constrainer-ffn-hidden-dim

Hidden dimension(s) in the bond constrainer FFN top model

Default: [300]

--bond-constrainer-ffn-num-layers

Number of layers in bond constrainer FFN top model

Default: 1

training input data args #

-w, --weight-column

Name of the column in the input CSV containing individual data weights

--target-columns

Name of the columns containing target values (by default, uses all columns except the SMILES column and the ignore_columns)

--mol-target-columns

Names of the columns containing mol target values (when training on mol and atom/bond targets simultaneously).

--atom-target-columns

Names of the columns containing atom target values.

--bond-target-columns

Names of the columns containing bond target values.

--ignore-columns

Name of the columns to ignore when target_columns is not provided

--no-cache

Turn off caching the featurized MolGraph s at the beginning of training

Default: False

--splits-column

Name of the column in the input CSV file containing ‘train’, ‘val’, or ‘test’ for each row.

training args #

-t, --task-type

Possible choices: regression, regression-mve, regression-evidential, regression-quantile, classification, classification-dirichlet, multiclass, multiclass-dirichlet, spectral

Type of dataset (determines the default loss function used during training, defaults to regression)

Default: 'regression'

-l, --loss-function

Possible choices: mse, mae, rmse, bounded-mse, bounded-mae, bounded-rmse, mve, evidential, quantile-point, pinball-point, bce, ce, binary-mcc, multiclass-mcc, dirichlet, sid, earthmovers, wasserstein, quantile, pinball, nlogprob_enrichment

Loss function to use during training (will use the default loss function for the given task type if not specified)

--v-kl, --evidential-regularization

Specify the value used in regularization for evidential loss function. The default value recommended by Soleimany et al. (2021) is 0.2. However, the optimal value is dataset-dependent, so it is recommended that users test different values to find the best value for their model.

Default: 0.0

--eps

Evidential regularization epsilon

Default: 1e-08

--alpha

Target error bounds for quantile interval loss

Default: 0.1

--metrics, --metric

Possible choices: mse, mae, rmse, bounded-mse, bounded-mae, bounded-rmse, r2, binary-mcc, multiclass-mcc, roc, prc, accuracy, f1

Specify the evaluation metrics. If unspecified, Chemprop will use the following metrics for given dataset types: regression -> rmse, classification -> roc, multiclass -> ce (‘cross entropy’), spectral -> sid. If multiple metrics are provided, the 0-th one will be used for early stopping and checkpointing.

--tracking-metric

The metric to track for early stopping, checkpointing, and hyperparameter optimization. Defaults to the criterion used during training. When training on two or three of molecule, atom, and bond targets, and not tracking the default (‘val_loss’), you must append ‘-mol’, ‘-atom’, or ‘-bond’ to the metric name to specify which individual metric to track. For example, ‘val_loss-bond’ will track the criterion value of the bond predictions and ‘rmse-atom’ will track the RMSE of the atom predictions.

Default: 'val_loss'

--show-individual-scores

Show all scores for individual targets, not just average, at the end.

Default: False

--task-weights

Weights to apply for whole tasks in the loss function

--warmup-epochs

Number of epochs during which learning rate increases linearly from init_lr to max_lr (afterwards, learning rate decreases exponentially from max_lr to final_lr)

Default: 2

--init-lr

Initial learning rate.

Default: 0.0001

--max-lr

Maximum learning rate.

Default: 0.001

--final-lr

Final learning rate.

Default: 0.0001

--epochs

Number of epochs to train over

Default: 50

--patience

Number of epochs to wait for improvement before early stopping

--min-delta

Minimum change in the monitored quantity to qualify as an improvement

Default: 0.0

--grad-clip

Passed directly to the lightning trainer which controls grad clipping (see the Trainer() docstring for details)

--class-balance

Ensures each training batch contains an equal number of positive and negative samples.

Default: False

split args #

--split, --split-type

Possible choices: SCAFFOLD_BALANCED, RANDOM_WITH_REPEATED_SMILES, RANDOM, KENNARD_STONE, KMEANS

Method of splitting the data into train/val/test (case insensitive)

Default: RANDOM

--split-sizes

Split proportions for train/validation/test sets

Default: [0.8, 0.1, 0.1]

--split-key-molecule

Specify the index of the key molecule used for splitting when multiple molecules are present and constrained split_type is used (e.g., scaffold_balanced or random_with_repeated_smiles). Note that this index begins with zero for the first molecule.

Default: 0

--num-replicates

Number of replicates.

Default: 1

-k, --num-folds

The -k/–num-folds argument was removed in v2.1.0 - use –num-replicates instead.

--save-smiles-splits

Whether to store the SMILES in each train/val/test split

Default: False

--save-data-splits

Whether to store the input data in each train/val/test split

Default: False

--splits-file

Path to a JSON file containing pre-defined splits for the input data, formatted as a list of dictionaries with keys train, val, and test and values as lists of indices or formatted strings (e.g. [0, 1, 2, 4] or ‘0-2,4’)

--data-seed

Specify the random seed to use when splitting data into train/val/test sets. When --num-replicates > 1, the first replicate uses this seed and all subsequent replicates add 1 to the seed (also used for shuffling data in build_dataloader when shuffle is True).

Default: 0

predict #

use a pretrained Chemprop model for prediction

chemprop predict [-h] [--logfile [LOGFILE]] [-v] [-q]
                 [-s SMILES_COLUMNS [SMILES_COLUMNS ...]]
                 [-r REACTION_COLUMNS [REACTION_COLUMNS ...]]
                 [--no-header-row] [-n NUM_WORKERS] [-b BATCH_SIZE]
                 [--accelerator ACCELERATOR] [--devices DEVICES]
                 [--rxn-mode {REAC_PROD,REAC_PROD_BALANCE,REAC_DIFF,REAC_DIFF_BALANCE,PROD_DIFF,PROD_DIFF_BALANCE}]
                 [--multi-hot-atom-featurizer-mode {V1,V2,ORGANIC,RIGR}]
                 [--keep-h] [--add-h] [--ignore-stereo] [--reorder-atoms]
                 [--molecule-featurizers {morgan_binary,morgan_count,rdkit_2d,v1_rdkit_2d,v1_rdkit_2d_normalized,charge} [{morgan_binary,morgan_count,rdkit_2d,v1_rdkit_2d,v1_rdkit_2d_normalized,charge} ...]]
                 [--descriptors-path DESCRIPTORS_PATH]
                 [--descriptors-columns DESCRIPTORS_COLUMNS [DESCRIPTORS_COLUMNS ...]]
                 [--no-descriptor-scaling] [--no-atom-feature-scaling]
                 [--no-atom-descriptor-scaling] [--no-bond-feature-scaling]
                 [--no-bond-descriptor-scaling]
                 [--atom-features-path ATOM_FEATURES_PATH [ATOM_FEATURES_PATH ...]]
                 [--atom-descriptors-path ATOM_DESCRIPTORS_PATH [ATOM_DESCRIPTORS_PATH ...]]
                 [--bond-features-path BOND_FEATURES_PATH [BOND_FEATURES_PATH ...]]
                 [--bond-descriptors-path BOND_DESCRIPTORS_PATH [BOND_DESCRIPTORS_PATH ...]]
                 [--constraints-path CONSTRAINTS_PATH]
                 [--constraints-to-targets CONSTRAINTS_TO_TARGETS [CONSTRAINTS_TO_TARGETS ...]]
                 [--use-cuikmolmaker-featurization] -i TEST_PATH [-o OUTPUT]
                 [--drop-extra-columns] --model-paths MODEL_PATHS
                 [MODEL_PATHS ...] [--cal-path CAL_PATH]
                 [--uncertainty-method {none,mve,ensemble,classification,evidential-total,evidential-epistemic,evidential-aleatoric,dropout,classification-dirichlet,multiclass-dirichlet,quantile-regression}]
                 [--calibration-method {zscaling,zelikman-interval,mve-weighting,conformal-regression,platt,isotonic,conformal-multilabel,conformal-multiclass,conformal-adaptive,isotonic-multiclass}]
                 [--evaluation-methods {nll-regression,miscalibration_area,ence,spearman,conformal-coverage-regression,nll-classification,conformal-coverage-classification,nll-multiclass,conformal-coverage-multiclass} [{nll-regression,miscalibration_area,ence,spearman,conformal-coverage-regression,nll-classification,conformal-coverage-classification,nll-multiclass,conformal-coverage-multiclass} ...]]
                 [--uncertainty-dropout-p UNCERTAINTY_DROPOUT_P]
                 [--dropout-sampling-size DROPOUT_SAMPLING_SIZE]
                 [--calibration-interval-percentile CALIBRATION_INTERVAL_PERCENTILE]
                 [--conformal-alpha CONFORMAL_ALPHA]
                 [--cal-descriptors-path CAL_DESCRIPTORS_PATH [CAL_DESCRIPTORS_PATH ...]]
                 [--cal-atom-features-path CAL_ATOM_FEATURES_PATH [CAL_ATOM_FEATURES_PATH ...]]
                 [--cal-atom-descriptors-path CAL_ATOM_DESCRIPTORS_PATH [CAL_ATOM_DESCRIPTORS_PATH ...]]
                 [--cal-bond-features-path CAL_BOND_FEATURES_PATH [CAL_BOND_FEATURES_PATH ...]]
                 [--cal-bond-descriptors-path CAL_BOND_DESCRIPTORS_PATH [CAL_BOND_DESCRIPTORS_PATH ...]]
                 [--cal-constraints-path CAL_CONSTRAINTS_PATH]

Named Arguments #

--logfile, --log

-v

Increase verbosity level to DEBUG

Default: False

-q

Decrease verbosity level to WARNING or ERROR if specified twice

Default: 0

--accelerator

Passed directly to the lightning Trainer()

Default: 'auto'

--devices

Passed directly to the lightning Trainer() (must be a single string of comma separated devices, e.g. ‘1, 2’ if specifying multiple devices)

Default: 'auto'

--constraints-path

--constraints-to-targets

The column names of the atom or bond targets that correspond to each constraint column in the constraints CSV.

-i, --test-path

Path to an input CSV file containing SMILES

-o, --output, --preds-path

Specify path to which predictions will be saved. If the file extension is .pkl, it will be saved as a pickle file. Otherwise, Chemprop will save predictions as a CSV. If multiple models are used to make predictions, the average predictions will be saved in the file, and another file ending in ‘_individual’ with the same file extension will save the predictions for each individual model, with the column names being the target names appended with the model index (e.g., ‘_model_<index>’).

--drop-extra-columns

Whether to drop all columns from the test data file besides the SMILES columns and the new prediction columns

Default: False

--model-paths, --model-path

Location of checkpoint(s) or model file(s) to use for prediction. It can be a path to either a single pretrained model checkpoint (.ckpt) or single pretrained model file (.pt), a directory that contains these files, or a list of path(s) and directory(s). If a directory, will recursively search and predict on all found (.pt) models.

--cal-constraints-path

Path to constraints applied to atomic/bond properties prediction for the calibration set.

Shared input data args #

-s, --smiles-columns

Column names in the input CSV containing SMILES strings (uses the 0th column by default)

-r, --reaction-columns

Column names in the input CSV containing reaction SMILES in the format REACTANT>AGENT>PRODUCT, where ‘AGENT’ is optional

--no-header-row

Turn off using the first row in the input CSV as column names

Default: False

--ignore-stereo

Ignore stereochemical information (R/S and Cis/Trans) in the input SMILES

Default: False

--reorder-atoms

Reorder atoms in the Chem.Mol object using the specified atom map numbers

Default: False

Dataloader args #

-n, --num-workers

Number of workers for parallel data loading where 0 means sequential (Warning: setting num_workers to a value greater than 0 can cause hangs on Windows and MacOS)

Default: 0

-b, --batch-size

Batch size

Default: 64

Featurization args #

--rxn-mode, --reaction-mode

Possible choices: REAC_PROD, REAC_PROD_BALANCE, REAC_DIFF, REAC_DIFF_BALANCE, PROD_DIFF, PROD_DIFF_BALANCE

Choices for construction of atom and bond features for reactions (case insensitive):

REAC_PROD: concatenates the reactants feature with the products feature
REAC_DIFF: concatenates the reactants feature with the difference in features between reactants and products (Default)
PROD_DIFF: concatenates the products feature with the difference in features between reactants and products
REAC_PROD_BALANCE: concatenates the reactants feature with the products feature, balances imbalanced reactions
REAC_DIFF_BALANCE: concatenates the reactants feature with the difference in features between reactants and products, balances imbalanced reactions
PROD_DIFF_BALANCE: concatenates the products feature with the difference in features between reactants and products, balances imbalanced reactions

Default: REAC_DIFF

--multi-hot-atom-featurizer-mode

Possible choices: V1, V2, ORGANIC, RIGR

Selects the multi-hot atom featurization scheme. This affects both non-reaction and reaction featurization (case insensitive):

V1: Corresponds to the original configuration employed in the Chemprop V1
V2: Tailored for a broad range of molecules, this configuration encompasses all elements in the first four rows of the periodic table, along with iodine. It is the default in Chemprop V2.
ORGANIC: This configuration is designed specifically for use with organic molecules for drug research and development and includes a subset of elements most common in organic chemistry, including H, B, C, N, O, F, Si, P, S, Cl, Br, and I.
RIGR: Modified V2 (default) featurizer using only the resonance-invariant atom and bond features.

Default: V2

--keep-h

Whether hydrogens explicitly specified in input should be kept in the mol graph

Default: False

--add-h

Whether hydrogens should be added to the mol graph

Default: False

--molecule-featurizers, --features-generators

Possible choices: morgan_binary, morgan_count, rdkit_2d, v1_rdkit_2d, v1_rdkit_2d_normalized, charge

Method(s) of generating molecule features to use as extra descriptors

--descriptors-path

Path to extra descriptors to concatenate to learned representation

--descriptors-columns

Column names in the input CSV containing extra datapoint descriptors, like temperature and pressure. See also –descriptors-path.

--no-descriptor-scaling

Turn off extra descriptor scaling

Default: False

--no-atom-feature-scaling

Turn off extra atom feature scaling

Default: False

--no-atom-descriptor-scaling

Turn off extra atom descriptor scaling

Default: False

--no-bond-feature-scaling

Turn off extra bond feature scaling

Default: False

--no-bond-descriptor-scaling

Turn off extra bond descriptor scaling

Default: False

--atom-features-path

--atom-descriptors-path

--bond-features-path

--bond-descriptors-path

--use-cuikmolmaker-featurization

Use cuik-molmaker package for accelerated atom and bond featurization.

Default: False

Uncertainty and calibration args #

--cal-path

Path to data file to be used for uncertainty calibration.

--uncertainty-method

Possible choices: none, mve, ensemble, classification, evidential-total, evidential-epistemic, evidential-aleatoric, dropout, classification-dirichlet, multiclass-dirichlet, quantile-regression

The method of calculating uncertainty.

Default: 'none'

--calibration-method

Possible choices: zscaling, zelikman-interval, mve-weighting, conformal-regression, platt, isotonic, conformal-multilabel, conformal-multiclass, conformal-adaptive, isotonic-multiclass

The method used for calibrating the uncertainty calculated with uncertainty method.

--evaluation-methods, --evaluation-method

Possible choices: nll-regression, miscalibration_area, ence, spearman, conformal-coverage-regression, nll-classification, conformal-coverage-classification, nll-multiclass, conformal-coverage-multiclass

The methods used for evaluating the uncertainty performance if the test data provided includes targets. Available methods are [nll, miscalibration_area, ence, spearman] or any available classification or multiclass metric.

--uncertainty-dropout-p

The probability to use for Monte Carlo dropout uncertainty estimation.

Default: 0.1

--dropout-sampling-size

The number of samples to use for Monte Carlo dropout uncertainty estimation. Distinct from the dropout used during training.

Default: 10

--calibration-interval-percentile

Sets the percentile used in the calibration methods. Must be in the range (1, 100).

Default: 95

--conformal-alpha

Target error rate for conformal prediction. Must be in the range (0, 1).

Default: 0.1

--cal-descriptors-path

Path to extra descriptors to concatenate to learned representation in calibration dataset.

--cal-atom-features-path

Path to the extra atom features in calibration dataset.

--cal-atom-descriptors-path

Path to the extra atom descriptors in calibration dataset.

--cal-bond-features-path

Path to the extra bond descriptors in calibration dataset.

--cal-bond-descriptors-path

Path to the extra bond descriptors in calibration dataset.

convert #

Convert model checkpoint (.pt) to more recent version (.pt).

chemprop convert [-h] [--logfile [LOGFILE]] [-v] [-q]
                 [-c {v1_to_v2,v2_0_to_v2_1}] -i INPUT_PATH [-o OUTPUT_PATH]

Named Arguments #

--logfile, --log

-v

Increase verbosity level to DEBUG

Default: False

-q

Decrease verbosity level to WARNING or ERROR if specified twice

Default: 0

-c, --conversion

Possible choices: v1_to_v2, v2_0_to_v2_1

Conversion to perform. Models converted from v1 to v2 must be run with the v1 featurizer via –multi-hot-atom-featurizer-mode v1.

Default: 'v1_to_v2'

-i, --input-path

Path to a model .pt checkpoint file

-o, --output-path

Path to which the converted model will be saved (CURRENT_DIRECTORY/STEM_OF_INPUT_newversion.pt by default)

fingerprint #

Use a pretrained Chemprop model to calculate learned representations.

chemprop fingerprint [-h] [--logfile [LOGFILE]] [-v] [-q]
                     [-s SMILES_COLUMNS [SMILES_COLUMNS ...]]
                     [-r REACTION_COLUMNS [REACTION_COLUMNS ...]]
                     [--no-header-row] [-n NUM_WORKERS] [-b BATCH_SIZE]
                     [--accelerator ACCELERATOR] [--devices DEVICES]
                     [--rxn-mode {REAC_PROD,REAC_PROD_BALANCE,REAC_DIFF,REAC_DIFF_BALANCE,PROD_DIFF,PROD_DIFF_BALANCE}]
                     [--multi-hot-atom-featurizer-mode {V1,V2,ORGANIC,RIGR}]
                     [--keep-h] [--add-h] [--ignore-stereo] [--reorder-atoms]
                     [--molecule-featurizers {morgan_binary,morgan_count,rdkit_2d,v1_rdkit_2d,v1_rdkit_2d_normalized,charge} [{morgan_binary,morgan_count,rdkit_2d,v1_rdkit_2d,v1_rdkit_2d_normalized,charge} ...]]
                     [--descriptors-path DESCRIPTORS_PATH]
                     [--descriptors-columns DESCRIPTORS_COLUMNS [DESCRIPTORS_COLUMNS ...]]
                     [--no-descriptor-scaling] [--no-atom-feature-scaling]
                     [--no-atom-descriptor-scaling]
                     [--no-bond-feature-scaling]
                     [--no-bond-descriptor-scaling]
                     [--atom-features-path ATOM_FEATURES_PATH [ATOM_FEATURES_PATH ...]]
                     [--atom-descriptors-path ATOM_DESCRIPTORS_PATH [ATOM_DESCRIPTORS_PATH ...]]
                     [--bond-features-path BOND_FEATURES_PATH [BOND_FEATURES_PATH ...]]
                     [--bond-descriptors-path BOND_DESCRIPTORS_PATH [BOND_DESCRIPTORS_PATH ...]]
                     [--constraints-path CONSTRAINTS_PATH]
                     [--constraints-to-targets CONSTRAINTS_TO_TARGETS [CONSTRAINTS_TO_TARGETS ...]]
                     [--use-cuikmolmaker-featurization] -i TEST_PATH
                     [-o OUTPUT] --model-paths MODEL_PATHS [MODEL_PATHS ...]
                     --ffn-block-index FFN_BLOCK_INDEX

Named Arguments #

--logfile, --log

-v

Increase verbosity level to DEBUG

Default: False

-q

Decrease verbosity level to WARNING or ERROR if specified twice

Default: 0

--accelerator

Passed directly to the lightning Trainer()

Default: 'auto'

--devices

Passed directly to the lightning Trainer() (must be a single string of comma separated devices, e.g. ‘1, 2’ if specifying multiple devices)

Default: 'auto'

--constraints-path

--constraints-to-targets

The column names of the atom or bond targets that correspond to each constraint column in the constraints CSV.

-i, --test-path

Path to an input CSV file containing SMILES

-o, --output, --preds-path

Specify the path where predictions will be saved. If the file extension is .npz, they will be saved as a npz file. Otherwise, the predictions will be saved as a CSV. The index of the model will be appended to the filename’s stem. By default, predictions will be saved to the same location as --test-path with ‘_fps’ appended (e.g., ‘PATH/TO/TEST_PATH_fps_0.csv’).

--model-paths, --model-path

Specify location of checkpoint(s) or model file(s) to use for prediction. It can be a path to either a single pretrained model checkpoint (.ckpt) or single pretrained model file (.pt), a directory that contains these files, or a list of path(s) and directory(s). If a directory, chemprop will recursively search and predict on all found (.pt) models.

--ffn-block-index

The index indicates which linear layer returns the encoding in the FFN. An index of 0 denotes the post-aggregation representation through a 0-layer MLP, while an index of 1 represents the output from the first linear layer in the FFN, and so forth.

Default: -1

Shared input data args #

-s, --smiles-columns

Column names in the input CSV containing SMILES strings (uses the 0th column by default)

-r, --reaction-columns

Column names in the input CSV containing reaction SMILES in the format REACTANT>AGENT>PRODUCT, where ‘AGENT’ is optional

--no-header-row

Turn off using the first row in the input CSV as column names

Default: False

--ignore-stereo

Ignore stereochemical information (R/S and Cis/Trans) in the input SMILES

Default: False

--reorder-atoms

Reorder atoms in the Chem.Mol object using the specified atom map numbers

Default: False

Dataloader args #

-n, --num-workers

Number of workers for parallel data loading where 0 means sequential (Warning: setting num_workers to a value greater than 0 can cause hangs on Windows and MacOS)

Default: 0

-b, --batch-size

Batch size

Default: 64

Featurization args #

--rxn-mode, --reaction-mode

Possible choices: REAC_PROD, REAC_PROD_BALANCE, REAC_DIFF, REAC_DIFF_BALANCE, PROD_DIFF, PROD_DIFF_BALANCE

Choices for construction of atom and bond features for reactions (case insensitive):

REAC_PROD: concatenates the reactants feature with the products feature
REAC_DIFF: concatenates the reactants feature with the difference in features between reactants and products (Default)
PROD_DIFF: concatenates the products feature with the difference in features between reactants and products
REAC_PROD_BALANCE: concatenates the reactants feature with the products feature, balances imbalanced reactions
REAC_DIFF_BALANCE: concatenates the reactants feature with the difference in features between reactants and products, balances imbalanced reactions
PROD_DIFF_BALANCE: concatenates the products feature with the difference in features between reactants and products, balances imbalanced reactions

Default: REAC_DIFF

--multi-hot-atom-featurizer-mode

Possible choices: V1, V2, ORGANIC, RIGR

Selects the multi-hot atom featurization scheme. This affects both non-reaction and reaction featurization (case insensitive):

V1: Corresponds to the original configuration employed in the Chemprop V1
V2: Tailored for a broad range of molecules, this configuration encompasses all elements in the first four rows of the periodic table, along with iodine. It is the default in Chemprop V2.
ORGANIC: This configuration is designed specifically for use with organic molecules for drug research and development and includes a subset of elements most common in organic chemistry, including H, B, C, N, O, F, Si, P, S, Cl, Br, and I.
RIGR: Modified V2 (default) featurizer using only the resonance-invariant atom and bond features.

Default: V2

--keep-h

Whether hydrogens explicitly specified in input should be kept in the mol graph

Default: False

--add-h

Whether hydrogens should be added to the mol graph

Default: False

--molecule-featurizers, --features-generators

Possible choices: morgan_binary, morgan_count, rdkit_2d, v1_rdkit_2d, v1_rdkit_2d_normalized, charge

Method(s) of generating molecule features to use as extra descriptors

--descriptors-path

Path to extra descriptors to concatenate to learned representation

--descriptors-columns

Column names in the input CSV containing extra datapoint descriptors, like temperature and pressure. See also –descriptors-path.

--no-descriptor-scaling

Turn off extra descriptor scaling

Default: False

--no-atom-feature-scaling

Turn off extra atom feature scaling

Default: False

--no-atom-descriptor-scaling

Turn off extra atom descriptor scaling

Default: False

--no-bond-feature-scaling

Turn off extra bond feature scaling

Default: False

--no-bond-descriptor-scaling

Turn off extra bond descriptor scaling

Default: False

--atom-features-path

--atom-descriptors-path

--bond-features-path

--bond-descriptors-path

--use-cuikmolmaker-featurization

Use cuik-molmaker package for accelerated atom and bond featurization.

Default: False

hpopt #

Perform hyperparameter optimization on the given task.

chemprop hpopt [-h] [--logfile [LOGFILE]] [-v] [-q]
               [-s SMILES_COLUMNS [SMILES_COLUMNS ...]]
               [-r REACTION_COLUMNS [REACTION_COLUMNS ...]] [--no-header-row]
               [-n NUM_WORKERS] [-b BATCH_SIZE] [--accelerator ACCELERATOR]
               [--devices DEVICES]
               [--rxn-mode {REAC_PROD,REAC_PROD_BALANCE,REAC_DIFF,REAC_DIFF_BALANCE,PROD_DIFF,PROD_DIFF_BALANCE}]
               [--multi-hot-atom-featurizer-mode {V1,V2,ORGANIC,RIGR}]
               [--keep-h] [--add-h] [--ignore-stereo] [--reorder-atoms]
               [--molecule-featurizers {morgan_binary,morgan_count,rdkit_2d,v1_rdkit_2d,v1_rdkit_2d_normalized,charge} [{morgan_binary,morgan_count,rdkit_2d,v1_rdkit_2d,v1_rdkit_2d_normalized,charge} ...]]
               [--descriptors-path DESCRIPTORS_PATH]
               [--descriptors-columns DESCRIPTORS_COLUMNS [DESCRIPTORS_COLUMNS ...]]
               [--no-descriptor-scaling] [--no-atom-feature-scaling]
               [--no-atom-descriptor-scaling] [--no-bond-feature-scaling]
               [--no-bond-descriptor-scaling]
               [--atom-features-path ATOM_FEATURES_PATH [ATOM_FEATURES_PATH ...]]
               [--atom-descriptors-path ATOM_DESCRIPTORS_PATH [ATOM_DESCRIPTORS_PATH ...]]
               [--bond-features-path BOND_FEATURES_PATH [BOND_FEATURES_PATH ...]]
               [--bond-descriptors-path BOND_DESCRIPTORS_PATH [BOND_DESCRIPTORS_PATH ...]]
               [--constraints-path CONSTRAINTS_PATH]
               [--constraints-to-targets CONSTRAINTS_TO_TARGETS [CONSTRAINTS_TO_TARGETS ...]]
               [--use-cuikmolmaker-featurization] [--config-path CONFIG_PATH]
               [-i [DATA_PATH ...]] [-o OUTPUT_DIR] [--remove-checkpoints]
               [--checkpoint CHECKPOINT [CHECKPOINT ...]] [--freeze-encoder]
               [--model-frzn MODEL_FRZN] [--frzn-ffn-layers FRZN_FFN_LAYERS]
               [--from-foundation FROM_FOUNDATION]
               [--ensemble-size ENSEMBLE_SIZE]
               [--message-hidden-dim MESSAGE_HIDDEN_DIM [MESSAGE_HIDDEN_DIM ...]]
               [--message-bias] [--depth DEPTH [DEPTH ...]] [--undirected]
               [--dropout DROPOUT] [--mpn-shared]
               [--aggregation {mean,sum,norm}]
               [--aggregation-norm AGGREGATION_NORM] [--atom-messages]
               [--activation {RELU,CELU,ELU,GELU,GLU,HARDSHRINK,HARDSIGMOID,HARDSWISH,HARDTANH,LEAKYRELU,LOGSIGMOID,LOGSOFTMAX,MISH,MULTIHEADATTENTION,PRELU,RRELU,RELU6,SILU,SIGMOID,SOFTMAX,SOFTMAX2D,SOFTMIN,SOFTPLUS,SOFTSHRINK,SOFTSIGN,TANH,TANHSHRINK,THRESHOLD}]
               [--activation-args [ACTIVATION_ARGS ...]]
               [--ffn-hidden-dim FFN_HIDDEN_DIM [FFN_HIDDEN_DIM ...]]
               [--ffn-num-layers FFN_NUM_LAYERS] [--batch-norm]
               [--multiclass-num-classes MULTICLASS_NUM_CLASSES]
               [--atom-task-weights ATOM_TASK_WEIGHTS [ATOM_TASK_WEIGHTS ...]]
               [--atom-ffn-hidden-dim ATOM_FFN_HIDDEN_DIM [ATOM_FFN_HIDDEN_DIM ...]]
               [--atom-ffn-num-layers ATOM_FFN_NUM_LAYERS]
               [--atom-multiclass-num-classes ATOM_MULTICLASS_NUM_CLASSES]
               [--bond-task-weights BOND_TASK_WEIGHTS [BOND_TASK_WEIGHTS ...]]
               [--bond-ffn-hidden-dim BOND_FFN_HIDDEN_DIM [BOND_FFN_HIDDEN_DIM ...]]
               [--bond-ffn-num-layers BOND_FFN_NUM_LAYERS]
               [--bond-multiclass-num-classes BOND_MULTICLASS_NUM_CLASSES]
               [--atom-constrainer-ffn-hidden-dim ATOM_CONSTRAINER_FFN_HIDDEN_DIM [ATOM_CONSTRAINER_FFN_HIDDEN_DIM ...]]
               [--atom-constrainer-ffn-num-layers ATOM_CONSTRAINER_FFN_NUM_LAYERS]
               [--bond-constrainer-ffn-hidden-dim BOND_CONSTRAINER_FFN_HIDDEN_DIM [BOND_CONSTRAINER_FFN_HIDDEN_DIM ...]]
               [--bond-constrainer-ffn-num-layers BOND_CONSTRAINER_FFN_NUM_LAYERS]
               [-w WEIGHT_COLUMN]
               [--target-columns TARGET_COLUMNS [TARGET_COLUMNS ...]]
               [--mol-target-columns MOL_TARGET_COLUMNS [MOL_TARGET_COLUMNS ...]]
               [--atom-target-columns ATOM_TARGET_COLUMNS [ATOM_TARGET_COLUMNS ...]]
               [--bond-target-columns BOND_TARGET_COLUMNS [BOND_TARGET_COLUMNS ...]]
               [--ignore-columns IGNORE_COLUMNS [IGNORE_COLUMNS ...]]
               [--no-cache] [--splits-column SPLITS_COLUMN]
               [-t {regression,regression-mve,regression-evidential,regression-quantile,classification,classification-dirichlet,multiclass,multiclass-dirichlet,spectral}]
               [-l {mse,mae,rmse,bounded-mse,bounded-mae,bounded-rmse,mve,evidential,quantile-point,pinball-point,bce,ce,binary-mcc,multiclass-mcc,dirichlet,sid,earthmovers,wasserstein,quantile,pinball,nlogprob_enrichment}]
               [--v-kl V_KL] [--eps EPS] [--alpha ALPHA]
               [--metrics {mse,mae,rmse,bounded-mse,bounded-mae,bounded-rmse,r2,binary-mcc,multiclass-mcc,roc,prc,accuracy,f1} [{mse,mae,rmse,bounded-mse,bounded-mae,bounded-rmse,r2,binary-mcc,multiclass-mcc,roc,prc,accuracy,f1} ...]]
               [--tracking-metric TRACKING_METRIC] [--show-individual-scores]
               [--task-weights TASK_WEIGHTS [TASK_WEIGHTS ...]]
               [--warmup-epochs WARMUP_EPOCHS] [--init-lr INIT_LR]
               [--max-lr MAX_LR] [--final-lr FINAL_LR] [--epochs EPOCHS]
               [--patience PATIENCE] [--min-delta MIN_DELTA]
               [--grad-clip GRAD_CLIP] [--class-balance]
               [--split {SCAFFOLD_BALANCED,RANDOM_WITH_REPEATED_SMILES,RANDOM,KENNARD_STONE,KMEANS}]
               [--split-sizes SPLIT_SIZES SPLIT_SIZES SPLIT_SIZES]
               [--split-key-molecule SPLIT_KEY_MOLECULE]
               [--num-replicates NUM_REPLICATES] [-k NUM_FOLDS]
               [--save-smiles-splits] [--save-data-splits]
               [--splits-file SPLITS_FILE] [--data-seed DATA_SEED]
               [--pytorch-seed PYTORCH_SEED]
               [--search-parameter-keywords SEARCH_PARAMETER_KEYWORDS [SEARCH_PARAMETER_KEYWORDS ...]]
               [--hpopt-save-dir HPOPT_SAVE_DIR]
               [--raytune-num-samples RAYTUNE_NUM_SAMPLES]
               [--raytune-search-algorithm {random,hyperopt,optuna}]
               [--raytune-trial-scheduler {FIFO,AsyncHyperBand}]
               [--raytune-num-workers RAYTUNE_NUM_WORKERS] [--raytune-use-gpu]
               [--raytune-num-checkpoints-to-keep RAYTUNE_NUM_CHECKPOINTS_TO_KEEP]
               [--raytune-grace-period RAYTUNE_GRACE_PERIOD]
               [--raytune-reduction-factor RAYTUNE_REDUCTION_FACTOR]
               [--raytune-temp-dir RAYTUNE_TEMP_DIR]
               [--raytune-num-cpus RAYTUNE_NUM_CPUS]
               [--raytune-num-gpus RAYTUNE_NUM_GPUS]
               [--raytune-max-concurrent-trials RAYTUNE_MAX_CONCURRENT_TRIALS]
               [--hyperopt-n-initial-points HYPEROPT_N_INITIAL_POINTS]
               [--hyperopt-random-state-seed HYPEROPT_RANDOM_STATE_SEED]

Named Arguments #

--logfile, --log

-v

Increase verbosity level to DEBUG

Default: False

-q

Decrease verbosity level to WARNING or ERROR if specified twice

Default: 0

--accelerator

Passed directly to the lightning Trainer()

Default: 'auto'

--devices

Passed directly to the lightning Trainer() (must be a single string of comma separated devices, e.g. ‘1, 2’ if specifying multiple devices)

Default: 'auto'

--constraints-path

--constraints-to-targets

The column names of the atom or bond targets that correspond to each constraint column in the constraints CSV.

--config-path

Path to a configuration file (command line arguments override values in the configuration file)

-i, --data-path

-o, --output-dir, --save-dir

Directory where training outputs will be saved (defaults to CURRENT_DIRECTORY/chemprop_training/STEM_OF_INPUT/TIME_STAMP)

--remove-checkpoints

Remove intermediate checkpoint files after training is complete.

Default: False

--ensemble-size

Number of models in ensemble for each splitting of data

Default: 1

--atom-multiclass-num-classes

Number of classes for atom targets when running multiclass classification

Default: 3

--bond-multiclass-num-classes

Number of classes for bond targets when running multiclass classification

Default: 3

--pytorch-seed

Seed for PyTorch randomness (e.g., random initial weights)

Shared input data args #

-s, --smiles-columns

Column names in the input CSV containing SMILES strings (uses the 0th column by default)

-r, --reaction-columns

Column names in the input CSV containing reaction SMILES in the format REACTANT>AGENT>PRODUCT, where ‘AGENT’ is optional

--no-header-row

Turn off using the first row in the input CSV as column names

Default: False

--ignore-stereo

Ignore stereochemical information (R/S and Cis/Trans) in the input SMILES

Default: False

--reorder-atoms

Reorder atoms in the Chem.Mol object using the specified atom map numbers

Default: False

Dataloader args #

-n, --num-workers

Number of workers for parallel data loading where 0 means sequential (Warning: setting num_workers to a value greater than 0 can cause hangs on Windows and MacOS)

Default: 0

-b, --batch-size

Batch size

Default: 64

Featurization args #

--rxn-mode, --reaction-mode

Possible choices: REAC_PROD, REAC_PROD_BALANCE, REAC_DIFF, REAC_DIFF_BALANCE, PROD_DIFF, PROD_DIFF_BALANCE

Choices for construction of atom and bond features for reactions (case insensitive):

REAC_PROD: concatenates the reactants feature with the products feature
REAC_DIFF: concatenates the reactants feature with the difference in features between reactants and products (Default)
PROD_DIFF: concatenates the products feature with the difference in features between reactants and products
REAC_PROD_BALANCE: concatenates the reactants feature with the products feature, balances imbalanced reactions
REAC_DIFF_BALANCE: concatenates the reactants feature with the difference in features between reactants and products, balances imbalanced reactions
PROD_DIFF_BALANCE: concatenates the products feature with the difference in features between reactants and products, balances imbalanced reactions

Default: REAC_DIFF

--multi-hot-atom-featurizer-mode

Possible choices: V1, V2, ORGANIC, RIGR

Selects the multi-hot atom featurization scheme. This affects both non-reaction and reaction featurization (case insensitive):

V1: Corresponds to the original configuration employed in the Chemprop V1
V2: Tailored for a broad range of molecules, this configuration encompasses all elements in the first four rows of the periodic table, along with iodine. It is the default in Chemprop V2.
ORGANIC: This configuration is designed specifically for use with organic molecules for drug research and development and includes a subset of elements most common in organic chemistry, including H, B, C, N, O, F, Si, P, S, Cl, Br, and I.
RIGR: Modified V2 (default) featurizer using only the resonance-invariant atom and bond features.

Default: V2

--keep-h

Whether hydrogens explicitly specified in input should be kept in the mol graph

Default: False

--add-h

Whether hydrogens should be added to the mol graph

Default: False

--molecule-featurizers, --features-generators

Possible choices: morgan_binary, morgan_count, rdkit_2d, v1_rdkit_2d, v1_rdkit_2d_normalized, charge

Method(s) of generating molecule features to use as extra descriptors

--descriptors-path

Path to extra descriptors to concatenate to learned representation

--descriptors-columns

Column names in the input CSV containing extra datapoint descriptors, like temperature and pressure. See also –descriptors-path.

--no-descriptor-scaling

Turn off extra descriptor scaling

Default: False

--no-atom-feature-scaling

Turn off extra atom feature scaling

Default: False

--no-atom-descriptor-scaling

Turn off extra atom descriptor scaling

Default: False

--no-bond-feature-scaling

Turn off extra bond feature scaling

Default: False

--no-bond-descriptor-scaling

Turn off extra bond descriptor scaling

Default: False

--atom-features-path

--atom-descriptors-path

--bond-features-path

--bond-descriptors-path

--use-cuikmolmaker-featurization

Use cuik-molmaker package for accelerated atom and bond featurization.

Default: False

transfer learning args #

--checkpoint

--freeze-encoder

Freeze the message passing layer from the checkpoint model (specified by --checkpoint).

Default: False

--model-frzn

Path to model checkpoint file to be loaded for overwriting and freezing weights. By default, all MPNN weights are frozen with this option.

--frzn-ffn-layers

Freeze the first n layers of the FFN from the checkpoint model (specified by --checkpoint). The message passing layer should also be frozen with --freeze-encoder.

Default: 0

--from-foundation

Name of pretrained foundation model used to initialize message passing. One of: CHEMELEON, or a path to a local model file.

message passing #

--message-hidden-dim

Hidden dimension of the messages (specify multiple values to customize multicomponent encoders separately)

Default: [300]

--message-bias

Add bias to the message passing layers

Default: False

--depth

Number of message passing steps (specify multiple values to customize multicomponent encoders separately)

Default: [3]

--undirected

Pass messages on undirected bonds/edges (always sum the two relevant bond vectors)

Default: False

--dropout

Dropout probability in message passing/FFN layers

Default: 0.0

--mpn-shared

Whether to use the same message passing neural network for all input molecules (only relevant if number_of_molecules > 1)

Default: False

--aggregation, --agg

Possible choices: mean, sum, norm

Aggregation mode to use during graph predictor

Default: 'norm'

--aggregation-norm

Normalization factor by which to divide summed up atomic features for norm aggregation

Default: 100

--atom-messages

Pass messages on atoms rather than bonds.

Default: False

--activation

Activation function in message passing/FFN layers.

Default: RELU

--activation-args

Arguments for the activation function (Example: arg1 arg2 key1=value1 key2=value2).

FFN args #

--ffn-hidden-dim

Hidden dimension(s) in the FFN top model. A single value is applied to all layers; multiple values specify per-layer widths (must match –ffn-num-layers)

Default: [300]

--ffn-num-layers

Number of hidden layers in FFN top model (v2 semantics, differs from v1)

Default: 1

extra MPNN args #

--batch-norm

Turn on batch normalization after aggregation

Default: False

--multiclass-num-classes

Number of classes when running multiclass classification

Default: 3

Atom FFN args #

--atom-task-weights

Weights to apply for all atom tasks in the loss function

--atom-ffn-hidden-dim

Hidden dimension(s) in the atom FFN top model

Default: [300]

--atom-ffn-num-layers

Number of layers in atom FFN top model

Default: 1

Bond FFN args #

--bond-task-weights

Weights to apply for all bond tasks in the loss function

--bond-ffn-hidden-dim

Hidden dimension(s) in the bond FFN top model

Default: [300]

--bond-ffn-num-layers

Number of layers in bond FFN top model

Default: 1

Atom constrainer FFN args #

--atom-constrainer-ffn-hidden-dim

Hidden dimension(s) in the atom constrainer FFN top model

Default: [300]

--atom-constrainer-ffn-num-layers

Number of layers in atom constrainer FFN top model

Default: 1

Bond constrainer FFN args #

--bond-constrainer-ffn-hidden-dim

Hidden dimension(s) in the bond constrainer FFN top model

Default: [300]

--bond-constrainer-ffn-num-layers

Number of layers in bond constrainer FFN top model

Default: 1

training input data args #

-w, --weight-column

Name of the column in the input CSV containing individual data weights

--target-columns

Name of the columns containing target values (by default, uses all columns except the SMILES column and the ignore_columns)

--mol-target-columns

Names of the columns containing mol target values (when training on mol and atom/bond targets simultaneously).

--atom-target-columns

Names of the columns containing atom target values.

--bond-target-columns

Names of the columns containing bond target values.

--ignore-columns

Name of the columns to ignore when target_columns is not provided

--no-cache

Turn off caching the featurized MolGraph s at the beginning of training

Default: False

--splits-column

Name of the column in the input CSV file containing ‘train’, ‘val’, or ‘test’ for each row.

training args #

-t, --task-type

Possible choices: regression, regression-mve, regression-evidential, regression-quantile, classification, classification-dirichlet, multiclass, multiclass-dirichlet, spectral

Type of dataset (determines the default loss function used during training, defaults to regression)

Default: 'regression'

-l, --loss-function

Loss function to use during training (will use the default loss function for the given task type if not specified)

--v-kl, --evidential-regularization

Default: 0.0

--eps

Evidential regularization epsilon

Default: 1e-08

--alpha

Target error bounds for quantile interval loss

Default: 0.1

--metrics, --metric

Possible choices: mse, mae, rmse, bounded-mse, bounded-mae, bounded-rmse, r2, binary-mcc, multiclass-mcc, roc, prc, accuracy, f1

--tracking-metric

Default: 'val_loss'

--show-individual-scores

Show all scores for individual targets, not just average, at the end.

Default: False

--task-weights

Weights to apply for whole tasks in the loss function

--warmup-epochs

Number of epochs during which learning rate increases linearly from init_lr to max_lr (afterwards, learning rate decreases exponentially from max_lr to final_lr)

Default: 2

--init-lr

Initial learning rate.

Default: 0.0001

--max-lr

Maximum learning rate.

Default: 0.001

--final-lr

Final learning rate.

Default: 0.0001

--epochs

Number of epochs to train over

Default: 50

--patience

Number of epochs to wait for improvement before early stopping

--min-delta

Minimum change in the monitored quantity to qualify as an improvement

Default: 0.0

--grad-clip

Passed directly to the lightning trainer which controls grad clipping (see the Trainer() docstring for details)

--class-balance

Ensures each training batch contains an equal number of positive and negative samples.

Default: False

split args #

--split, --split-type

Possible choices: SCAFFOLD_BALANCED, RANDOM_WITH_REPEATED_SMILES, RANDOM, KENNARD_STONE, KMEANS

Method of splitting the data into train/val/test (case insensitive)

Default: RANDOM

--split-sizes

Split proportions for train/validation/test sets

Default: [0.8, 0.1, 0.1]

--split-key-molecule

Default: 0

--num-replicates

Number of replicates.

Default: 1

-k, --num-folds

The -k/–num-folds argument was removed in v2.1.0 - use –num-replicates instead.

--save-smiles-splits

Whether to store the SMILES in each train/val/test split

Default: False

--save-data-splits

Whether to store the input data in each train/val/test split

Default: False

--splits-file

--data-seed

Default: 0

Chemprop hyperparameter optimization arguments #

--search-parameter-keywords

The model parameters over which to search for an optimal hyperparameter configuration. Some options are bundles of parameters or otherwise special parameter operations. Special keywords include:

basic: Default set of hyperparameters for search (depth, ffn_num_layers, dropout, message_hidden_dim, and ffn_hidden_dim)

learning_rate: Search for max_lr, init_lr_ratio, final_lr_ratio, and warmup_epochs. The search for init_lr and final_lr values are defined as fractions of the max_lr value. The search for warmup_epochs is as a fraction of the total epochs used.

all: Include search for all 13 individual keyword options (including: activation, aggregation, aggregation_norm, and batch_size which aren’t included in the other two keywords).

Individual supported parameters:: [‘activation’, ‘dropout’, ‘message_hidden_dim’, ‘depth’, ‘aggregation’, ‘aggregation_norm’, ‘ffn_hidden_dim’, ‘ffn_num_layers’, ‘atom_ffn_hidden_dim’, ‘atom_ffn_num_layers’, ‘atom_constrainer_ffn_hidden_dim’, ‘atom_constrainer_ffn_num_layers’, ‘bond_ffn_hidden_dim’, ‘bond_ffn_num_layers’, ‘bond_constrainer_ffn_hidden_dim’, ‘bond_constrainer_ffn_num_layers’, ‘batch_size’, ‘init_lr_ratio’, ‘max_lr’, ‘final_lr_ratio’, ‘warmup_epochs’]

Default: ['basic']

--hpopt-save-dir

Directory to save the hyperparameter optimization results

Ray Tune arguments #

--raytune-num-samples

Passed directly to Ray Tune TuneConfig to control number of trials to run

Default: 10

--raytune-search-algorithm

Possible choices: random, hyperopt, optuna

Passed to Ray Tune TuneConfig to control search algorithm

Default: 'hyperopt'

--raytune-trial-scheduler

Possible choices: FIFO, AsyncHyperBand

Passed to Ray Tune TuneConfig to control trial scheduler

Default: 'FIFO'

--raytune-num-workers

Passed directly to Ray Tune ScalingConfig to control number of workers to use

Default: 1

--raytune-use-gpu

Passed directly to Ray Tune ScalingConfig to control whether to use GPUs

Default: False

--raytune-num-checkpoints-to-keep

Passed directly to Ray Tune CheckpointConfig to control number of checkpoints to keep

Default: 1

--raytune-grace-period

Passed directly to Ray Tune ASHAScheduler to control grace period

Default: 10

--raytune-reduction-factor

Passed directly to Ray Tune ASHAScheduler to control reduction factor

Default: 2

--raytune-temp-dir

Passed directly to Ray Tune init to control temporary directory

--raytune-num-cpus

Passed directly to Ray Tune init to control number of CPUs to use

--raytune-num-gpus

Passed directly to Ray Tune init to control number of GPUs to use

--raytune-max-concurrent-trials

Passed directly to Ray Tune TuneConfig to control maximum concurrent trials

Hyperopt arguments #

--hyperopt-n-initial-points: Passed directly to HyperOptSearch to control number of initial points to sample
--hyperopt-random-state-seed: Passed directly to HyperOptSearch to control random state seed

CLI Reference

Contents

CLI Reference#