Table 2.

Machine learning models for antibiotic discovery.

	Public release
Algorithm	Code	Data	Software	Software type
Antimicrobial activity prediction
Artificial neural network40		Yes
Support vector machine38		Yes
Multinomial logistic regression33		Yes
LSTM RNN44	Yes	Yes	Yes	Command-line tool
XGBoost42	Yes	Yes	Yes	Command-line tool
Directed-message passing neural network16	Yes	Yes	Yes	Web server, Docker container
DBSCAN47		Yes	Yes	Web server
DBSCAN48			Yes	Web server
Convolutional neural network41		Yes	Yes	Web server
Generalized linear model49
Random forest50
Hemolytic activity prediction
Classification and regression trees55		Yes
Artificial neural network54		Yes	Yes	Web server
Gradient boosting classifiers56	Yes	Yes
Support vector machine183		Yes	Yes	Web server, mobile app, standalone
De novo antibiotic design
Variational autoencoder45		Yes
LSTM RNN30	Yes	Yes	Yes	Command-line tool
LSTM RNN120
Generative adversarial network119	Yes	Yes	Yes	Command-line tool

Machine learning models cited in this review pertain specifically to antimicrobial compound discovery, i.e., those that predict antimicrobial activity, those trained on antimicrobial compound data to predict drug-likeness, and those that generate potential antimicrobials. Public release of model source code, training and/or testing data, and/or associated software tools are noted. Criteria for data release were lenient, with “yes” indicating partial or full release of training or testing data.