Selected Publications

Preprint. Large-scale benchmarks to assess models for RNA fitness & structure prediction.

Science, 2025. A microfluidic platform to measure catalytic constants for hundreds of Adenylate Kinase variants, enabling mapping of the sequence-catalysis landscape and development of superior predictive models.

ICML, 2025. an end-to-end differentiable framework that learns to retrieve relevant protein homologs while simultaneously training for downstream tasks, achieving state-of-the-art fitness prediction performance while being orders of magnitude faster than traditional MSA-based approaches.

Preprint. A multiplexed experimental approach to analyze conformational fluctuations across thousands of protein domains, revealing hidden variations that affect protein cooperativity and function.

NeurIPS, 2024. A multimodal framework that integrates protein sequence, structure, and surface topology features to achieve state-of-the-art fitness prediction.

Preprint. A conditional autoregressive transformer model trained on 14.6 million mammalian promoter sequences that achieves state-of-the-art performance in predicting the effects of indels in human promoter regions.

Nature Biotech, 2024. A unifying framework that classifies models on the basis of their use of three core data modalities: sequences, structures and functional labels is introduced to make sense of the exploding diversity of machine learning approaches.

NeurIPS, 2023. Large-scale benchmarks to assess models for protein fitness prediction and design.

NeurIPS, 2023. A conditional semi-supervised pseudo-generative model for fitness prediction and design.

Nature, 2023. A computational framework to predict viral escape from pre-pandemic information only (evolutionary data and 3D structure).

ICML, 2023. A sample-efficient method for discovering optimal sets that are both diverse and optimize the function of interest.

NeurIPS, LMRL, 2022. A hybrid family-specific and family-agnostic model to achieve SOTA performance on protein fitness prediction and human variant annotation.

ICML, 2022. A suite of autoregressive transformers with biological priors, augmented with inference-time retrieval, to achieve SOTA performance on protein fitness prediction.

ICML, WCB, 2022. The first paper investigating scaling laws in protein language modeling.

Nature, 2021. Deep Generative Models (Bayesian VAEs) of evolutionary sequences to predict the effects of missense mutations in human proteins.

ICLR, 2022. A benchmark suite for evaluating active learning algorithms for experimental design in drug discovery.

NeurIPS, 2021. A framework that uses the epistemic uncertainty of the decoder of a VAE to guide the optimization of properties of high-dimensional structured objects (e.g., molecules) in latent space.

Preprint. A memory-efficient dropout-inspired scheme to train large neural networks faster with no loss in accuracy.