#### Antibiotics

A model trained to predict the probability that a molecule will inhibit the growth of E. coli. This model was trained on the data from A Deep Learning Approach to Antibiotic Discovery (processed data is here). The model is an ensemble of 20 chemprop models augmented with RDKit features and with optimized hyperparamers.

#### SARS

A model trained to predict the probability that a molecule will inhibit the 3CL protease of SARS-CoV. This model was trained on the data from PubChem assay AID1706 (processed data is here). The model is an ensemble of 5 chemprop models augmented with RDKit features.

#### SARS - Balanced

The same as SARS but trained using a class balance approach where each training batch samples an equal number of positive and negative molecules. Due to the extreme class imbalance of AID1706 (0.1% positive, 99.9% negative), this class balance approach encourages the model to produce predictions in a more reasonable range rather than predicting extremely low probabilities for almost all molecules (due to the weight of the negatives).

#### AMU SARS-CoV-2 *in vitro*

A model trained to predict the probability that a molecule will inhibit SARS-CoV-2 replication *in vitro*. This model was trained on data from *In vitro* screening of a FDA approved chemical library reveals potential inhibitors of SARS-CoV-2 replication (processed data is here). The model is an ensemble of 5 chemprop models augmented with RDKit features and with optimized hyperparameters.