Surojit Biswas

This paper introduces RGN2, a deep learning model that predicts protein structure from single sequences without MSAs, outperforming AlphaFold2 and RoseTTAFold on orphan proteins and achieving faster computation.

This paper introduces a data-efficient deep learning approach for protein engineering that uses minimal experimental data to predict and optimize protein function with high throughput.

This paper introduces UniRep, a deep learning-based representation of protein sequences that enables accurate prediction of protein stability and function while reducing engineering costs.

This paper explores using supervised learning with DNA synthesis and high-throughput screening to improve protein design, demonstrating generalization to unseen fitness landscape regions.

Tradict predicts transcriptome-wide gene abundances and transcriptional programs using 100 marker genes, showing robustness and accuracy across 23,000 RNA-Seq datasets.