Fine-tuning Protein Language Models with Deep Mutational Scanning improves Variant Effect Prediction

Protein Language Models (PLMs) have emerged as performant and scalable toolsfor predicting the functional impact and clinical significance of protein-codingvariants, but they still lag experimental accuracy. Here, we present a novel finetuningapproach to improve the performance of PLMs with experimental maps ofvariant effects from Deep Mutational Scanning (DMS) assays using a NormalisedLog-odds Ratio (NLR) head. We find consistent improvements in a held-out proteintest set, and on independent DMS and clinical variant annotation benchmarksfrom ProteinGym and ClinVar. These findings demonstrate that DMS is a promisingsource of sequence diversity and supervised training data for improving theperformance of PLMs for variant effect prediction.