Alex Hawkins‐Hooker

ProFam is an open-source protein family language model that improves fitness prediction and design by leveraging evolutionary context through homologous sequences.

This study benchmarks zero-shot single-cell foundation model embeddings for perturbation effect prediction and finds they do not consistently improve over baseline models under distribution shift.

This paper compares likelihood-based and embedding-based methods for fitness prediction and design using protein language models, finding likelihood-based fine-tuning consistently outperforms embedding-based approaches.

Chainsaw uses a fully convolutional neural network to predict protein domain boundaries, achieving higher accuracy than existing methods in domain segmentation.

The paper introduces eDICE, a deep learning model for imputing epigenomic tracks, showing improved performance on reference data and successful application to individual-specific epigenomic patterns.

The paper evaluates 23 imputation models for cross-cell type epigenomic data, finding quality assessment challenging due to distributional shifts, data availability, and redundant metrics.

This paper uses variational autoencoders to generate functional protein variants of bacterial luciferase, showing high success rates in retaining activity despite significant sequence changes.

The paper introduces a projection layer to improve deep learning models for regulatory DNA function prediction, showing performance gains on the DeepSEA dataset.