Juannan Zhou

The paper shows that synthesis cost minimization drives convergent amino acid changes in plastid ribosomal proteins across plant lineages with relaxed selection.

The paper establishes the relationship between Gaussian process priors and regularized regression in sequence-to-function mapping, showing how to construct regularizers that correspond to explicit Gaussian process priors and characterize implicit function space priors.

This study introduces pVEP, a framework showing how genetic background affects AI predictions of variant effects, revealing variant effects vary across haplotypes.

This study tracks mutation rates and fitness effects in C. elegans under minimal selection, finding increased indel mutation rates and no change in nucleotide substitution rates.

BacPT, a proteome foundation model, improves predictions of bacterial enzyme activities, biosynthetic clusters, metabolic traits, and ecological interactions using unsupervised deep learning.

The paper introduces interpretable Gaussian process models for sequence-function relationships, capturing epistasis and achieving superior predictive performance on large biological datasets.

This paper introduces a family of gauges for sequence-function models to remove gauge freedoms and improve model interpretation.

This paper introduces a whole bacterial proteome foundation model trained on ESM embeddings to capture gene interactions and genomic integrity, achieving state-of-the-art results in operon identification and genotype-phenotype mapping.

gpmap-tools uses Gaussian processes to infer and visualize genotype-phenotype maps with higher-order epistasis; across 262,144 Shine-Dalgarno variants it reveals register-specific peaks linked by three-nucleotide-offset ridges.

This study introduces an interpretable neural network to analyze higher-order epistasis in protein sequence-function relationships, finding that higher-order interactions contribute up to 60% of the epistatic variance.

This study infers the distribution of fitness effects of mutations in C. elegans using recombinant inbred lines, finding that positive effects are common but LD can skew results.

This paper introduces a differentiable Smith-Waterman algorithm to jointly learn multiple sequence alignments and downstream tasks, showing improvements in contact prediction and structure prediction.

The paper introduces a non-parametric Bayesian method for density estimation in sequence space and applies it to study pre-mRNA splicing and karyotypic diversity in human cancer.

The paper introduces a method to impute sequence-function relationships by minimizing epistasis, showing it provides a less epistatic fit with comparable predictive power.