Abstract: Systems and methods for generative language model tuning can include training the generative language model to generate sets of output text tokens with set of intermediary text tokens with training examples that include input and output pairs. The training can include processing the input with the language model to determine a predicted output and a predicted set of intermediary text tokens. The predicted set of intermediary text tokens can then be evaluated based at least in part on the output associated with the input and the predicted output.

Abstract: Provided herein are improved methods for training graph neural networks (GNNs) to predict the binding affinity of novel compounds based on data generated from DNA-encoded library (DEL) experiments. These methods include training the GNN to predict affinity for the target directly and then applying the predicted affinity to a model of the DEL experiment process to generate a predicted DEL read count. The predicted DEL read count can then be compared to an experimentally-observed DEL, read count to generate a loss value. The loss value can then be used to update the GNN as part of a GNN training process. The loss value can be augmented with simulated disynthon data generated from the predicted affinity.

Abstract: Systems and methods for generative language model tuning can include training the generative language model to generate sets of output text tokens with set of intermediary text tokens with training examples that include input and output pairs. The training can include processing the input with the language model to determine a predicted output and a predicted set of intermediary text tokens. The predicted set of intermediary text tokens can then be evaluated based at least in part on the output associated with the input and the predicted output.