K. K. Yang

Feature Reuse and Scaling: Understanding Transfer Learning with Protein Language Models

This study analyzes transfer learning in protein language models, finding that downstream task performance often relies on early pretraining features rather than scaling with pretraining duration.

Protein generation with evolutionary diffusion: sequence is all you need

This paper introduces EvoDiff, a diffusion model that generates diverse, high-fidelity proteins in sequence space, overcoming limitations of structure-based models.

Benchmarking Uncertainty Quantification for Protein Engineering

This study benchmarks deep learning uncertainty quantification methods on protein sequence-function regression tasks to improve biological sequence design.

Computational Scoring and Experimental Evaluation of Enzymes Generated by Neural Networks

The paper evaluates computational metrics for predicting enzyme activity in sequences generated by three models and finds that existing metrics like sequence identity and AlphaFold2 scores are not predictive.

Masked inverse folding with sequence transfer for protein representation learning

This study introduces a masked inverse folding model for protein representation learning, combining sequence and structural information to improve performance on protein engineering tasks.

Convolutions are competitive with transformers for protein sequence pretraining

This study shows that convolutional neural networks can be as effective as transformers for protein sequence pretraining, offering better scalability and performance on long sequences.

Randomized gates eliminate bias in sort-seq assays

The paper introduces randomized sorting in sort-seq assays to eliminate systematic bias, validated through simulation and experiments, and extends the method for variance estimation and multi-bin sorting.

FLIP: Benchmark tasks in fitness landscape inference for proteins

FLIP introduces a benchmark for fitness landscape inference in proteins, focusing on function prediction and model generalization in protein engineering scenarios.

A Topological Data Analytic Approach for Discovering Biophysical Signatures in Protein Dynamics

SINATRA Pro identifies topological differences between protein structures from molecular dynamics simulations, revealing both known and novel biophysical signatures with improved resolution.

Machine learning-guided channelrhodopsin engineering enables minimally-invasive optogenetics

The paper engineers light-gated channelrhodopsins with high photocurrent and light sensitivity, enabling minimally-invasive optogenetics without invasive surgery or implants.