Ahmed Elnaggar

This study benchmarks single-cell foundation models for data integration using real-world RNA-seq data and finds that fine-tuned scGPT_CP performs best in technical correction while scVI is the top overall performer.

Protein-specific tuning of masking, depth and positional embeddings, rather than more parameters, yields an encoder-decoder model that tops structure and function benchmarks at a fraction of the compute.

This paper trains and evaluates language models on protein sequences to predict structural and localization features, achieving state-of-the-art results without evolutionary information.

This paper introduces an end-to-end multitask learning approach using language models to predict protein structure and function without evolutionary alignments, achieving faster predictions with moderate accuracy.

This paper introduces SeqVec, a method to represent protein sequences as continuous vectors using deep learning, achieving better prediction results than methods relying on evolutionary information.