Ava P. Amini

Protein language models and the long tail of functional diversity

The paper examines the role of singleton protein sequences in training protein language models, finding they contain valuable functional diversity and should be included in training.

FLIP2: Expanding Protein Fitness Landscape Benchmarks for Real-World Machine Learning Applications

FLIP2 adds seven protein fitness datasets and 16 splits for enzymes, light-sensitive proteins, and interactions; simple ridge and zero-shot baselines often rival fine-tuned protein language models.

Evaluating the role of pre-training dataset size and diversity on single-cell foundation model performance

This study evaluates how pre-training dataset size and diversity affect single-cell foundation model performance on zero-shot and fine-tuned tasks using 22.2 million cells.

Beyond alignment: synergistic integration is required for multimodal cell foundation models

This paper introduces a metric to evaluate synergistic integration in multimodal cell models and shows when it benefits tasks over unimodal approaches.

Scalable nonparametric clustering with unified marker gene selection for single-cell RNA-seq data

This paper introduces NCLUSION, a nonparametric clustering method for scRNA-seq that identifies marker genes and clusters simultaneously, showing improved performance and reduced runtime.

Trainable subnetworks reveal insights into structure knowledge organization in protein language models

This study introduces trainable subnetworks to analyze how protein language models factorize structural information and shows that structure prediction is sensitive to these factorized representations.

Adaptive resampling for improved machine learning in imbalanced single-cell datasets

This paper introduces an adaptive resampling method to improve machine learning performance on imbalanced single-cell datasets, showing better results in gene expression reconstruction, cell type classification, and perturbation response prediction.

The Dayhoff Atlas: scaling sequence diversity for improved protein generation

The paper introduces the Dayhoff Atlas, a resource with large protein sequence datasets and generative models to improve protein design and prediction.

Hierarchical cross-entropy loss improves atlas-scale single-cell annotation models

This paper introduces a hierarchical cross-entropy loss to improve single-cell annotation models by aligning them with biological hierarchies, achieving better out-of-distribution performance.

ProtNote: a multimodal method for protein-function annotation

ProtNote is a multimodal deep learning model that enables supervised and zero-shot protein function prediction using free-form text inputs, generalizing to unseen functions.

A knockoff calibration method to avoid over-clustering in single-cell RNA-sequencing

This paper introduces a method to prevent over-clustering in scRNA-seq by controlling double-dipping, improving differential expression analysis accuracy.

Consequences of training data composition for deep learning models in single-cell biology

This study examines how training data composition affects deep learning models in single-cell transcriptomics, finding poor generalization to new cell types and benefits from including embryonic stem cell differentiation data.

Mutation and cell state compatibility is required and targetable in Ph+ acute lymphoblastic leukemia minimal residual disease

This study identifies mutation and transcriptional state compatibility in Ph+ ALL MRD, showing that targeting these interactions can overcome BCR::ABL1 independence and improve MRD monitoring.

Assessing the limits of zero-shot foundation models in single-cell biology

Used without fine-tuning, Geneformer and scGPT show limited reliability for cell type clustering and batch effect correction, often landing behind simpler methods.

Deep self-supervised learning for biosynthetic gene cluster detection and product classification

This paper introduces a self-supervised learning approach to detect and classify biosynthetic gene clusters in microbial genomes using masked language models trained on protein domains.

A nanoparticle priming agent reduces cellular uptake of cell-free DNA and enhances the sensitivity of liquid biopsies

A nanoparticle priming agent increases cfDNA levels in blood, improving ctDNA detection sensitivity in tumor-bearing mice.