Biology subjects

Aghazadeh, A.

Publications and source records attributed to Aghazadeh, A..

5 recordsLinked to original sources

bioRxiv2026

Protein fitness landscapes are simpler under evolutionary distributions

Understanding how mutations combine to shape protein fitness remains a central challenge in biology, driven in part by the prevalence of highorder epistasis. Existing analyses of epistasis, however, implicitly define epistatic interactions under a uniform probability measure over sequence space, even though evolution constrains natural proteins to a highly structured, non-uniform distribution of sequences. Here, we show that the apparent complexity of protein epistasis depends fundamentally on the underlying evolutionary distribution of sequences. We develop an evolution-aware spectral framework that incorporates the evolutionary distribution of amino acids at each sequence position, inducing an orthogonal decomposition under the evolutionary measure while preserving efficient spectral algorithms for scalable analysis. Across diverse protein fitness landscapes, this framework consistently produces more compact spectral representations, explaining more phenotypic variation with fewer epistatic interactions while substantially reducing apparent high-order epistasis. It also enables more accurate recovery of fitness landscapes from limited experimental measurements and concentrates the remaining higher-order interactions into localized, structurally interpretable motifs. These results suggest that a substantial fraction of apparent high-order epistasis arises from defining epistatic interactions under a uniform measure over sequence space and can be resolved by aligning spectral analysis with evolutionary constraints.

evolutionary biology↗

bioRxiv2025

GOLF: A Generative AI Framework for Pathogenicity Prediction of Myocilin OLF Variants

Missense mutations in the MYOC gene, particularly those affecting the olfactomedin (OLF) domain of the myocilin protein, can be causal for open-angle glaucoma--a leading cause of irre-versible blindness. However, predicting the pathogenicity of these mutations remains challenging due to the complex effects of toxic gain-of-function variants and the scarcity of labeled clinical data. Herein, we present GOLF, a generative AI framework for assessing and explaining the pathogenicity of OLF domain variants. GOLF collects and curates a comprehensive dataset of OLF homologs and trains generative models that predict the effect of monoallelic missense mutations. While these models exhibit diverse predictive behaviors, they collectively achieve accurate classification of known pathogenic and benign variants. To interpret their decision mechanisms, GOLF uses a sparse autoencoder (SAE) that reveals the underlying biochemical features exploited by the generative models to predict variant effects. GOLF enables accurate evaluation of disease-causing mutations, supporting early genetic risk stratification for glaucoma and facilitating interpretable investigations into the molecular basis of pathogenic variants.

bioinformatics↗

bioRxiv2021

On the sparsity of fitness functions and implications for learning

Fitness functions map biological sequences to a scalar property of interest. Accurate estimation of these functions yields biological insight and sets the foundation for model-based sequence design. However, the amount of fitness data available to learn these functions is typically small relative to the large combinatorial space of sequences; characterizing how much data is needed for accurate estimation remains an open problem. There is a growing body of evidence demonstrating that empirical fitness functions display substantial sparsity when represented in terms of epistatic interactions. Moreover, the theory of Compressed Sensing provides scaling laws for the number of samples required to exactly recover a sparse function. Motivated by these results, we develop a framework to study the sparsity of fitness functions sampled from a generalization of the NK model, a widely-used random field model of fitness functions. In particular, we present results that allow us to test the effect of the Generalized NK (GNK) models interpretable parameters--sequence length, alphabet size, and assumed interactions between sequence positions--on the sparsity of fitness functions sampled from the model and, consequently, the number of measurements required to exactly recover these functions. We validate our framework by demonstrating that GNK models with parameters set according to structural considerations can be used to accurately approximate the number of samples required to recover two empirical protein fitness functions and an RNA fitness function. In addition, we show that these GNK models identify important higher-order epistatic interactions in the empirical fitness functions using only structural information.

bioinformatics↗

bioRxiv2020

Sparse Epistatic Regularization of Deep Neural Networks for Inferring Fitness Functions

Despite recent advances in high-throughput combinatorial mutagenesis assays, the number of labeled sequences available to predict molecular functions has remained small for the vastness of the sequence space combined with the ruggedness of many fitness functions. Expressive models in machine learning (ML), such as deep neural networks (DNNs), can model the nonlinearities in rugged fitness functions, which manifest as high-order epistatic interactions among the mutational sites. However, in the absence of an inductive bias, DNNs overfit to the small number of labeled sequences available for training. Herein, we exploit the recent biological evidence that epistatic interactions in many fitness functions are sparse; this knowledge can be used as an inductive bias to regularize DNNs. We have developed a method for sparse epistatic regularization of DNNs, called the epistatic net (EN), which constrains the number of non-zero coefficients in the spectral representation of DNNs. For larger sequences, where finding the spectral transform becomes computationally intractable, we have developed a scalable extension of EN, which subsamples the combinatorial sequence space uniformly inducing a sparse-graph-code structure, and regularizes DNNs using the resulting greedy optimization method. Results on several biological landscapes, from bacterial to protein fitness functions, show that EN consistently improves the prediction accuracy of DNNs and enables them to outperform competing models which assume other forms of inductive biases. EN estimates all the higher-order epistatic interactions of DNNs trained on massive sequence spaces--a computational problem that takes years to solve without leveraging the epistatic sparsity in the fitness functions.

bioinformatics↗

bioRxiv2018

Systematic characterization of genome editing in primary T cells reveals proximal genomic insertions and enables machine learning prediction of CRISPR-Cas9 DNA repair outcomes

The Streptococcus pyogenes Cas9 (SpCas9) nuclease has become a ubiquitous genome editing tool due to its ability to target almost any location in DNA and create a double-stranded break1,2. After DNA cleavage, the break is fixed with endogenous DNA repair machinery, either by non-templated mechanisms (e.g. non-homologous end joining (NHEJ) or microhomology-mediated end joining (MMEJ)), or homology directed repair (HDR) using a complementary template sequence3,4. Previous work has shown that the distribution of repair outcomes within a cell population is non-random and dependent on the targeted sequence, and only recent efforts have begun to investigate this further5-11. However, no systematic work to date has been validated in primary human cells5,7. Here, we report DNA repair outcomes from 1,521 unique genomic locations edited with SpCas9 ribonucleoprotein complexes (RNPs) in primary human CD4+ T cells isolated from multiple healthy blood donors. We used targeted deep sequencing to measure the frequency distribution of repair outcomes for each guide RNA and discovered distinct features that drive individual repair outcomes after SpCas9 cleavage. Predictive features were combined into a new machine learning model, CRISPR Repair OUTcome (SPROUT), that predicts the length and probability of nucleotide insertions and deletions with R2 greater than 0.5. Surprisingly, we also observed large insertions at more than 90% of targeted loci, albeit at a low frequency. The inserted sequences aligned to diverse regions in the genome, and are enriched for sequences that are physically proximal to the break site due to chromatin interactions. This suggests a new mechanism where sequences from three-dimensionally neighboring regions of the genome can be inserted during DNA repair after Cas9-induced DNA breaks. Together, these findings provide powerful new predictive tools for Cas9-dependent genome editing and reveal new outcomes that can result from genome editing in primary T cells.

cell biology↗