Biology subjects

Mofrad, M.

Publications and source records attributed to Mofrad, M..

15 recordsLinked to original sources

bioRxiv2026

Decision-relevant measurement for precision microbiome interventions

Human gut microbiome studies often rely on stool, despite increasing evidence that it incompletely represents the spatially heterogeneous intestinal ecosystem. Whether this loss of spatial detail matters for downstream intervention decisions is largely unresolved. We treat compositional reconstruction and decision preservation as separate objectives. Using 16 matched recovery events from 11 participants, each containing stool and four intended regional capsule measurements, we aggregated amplicon sequence variants into 11 prespecified taxonomic and functional proxy guilds. We evaluated models that reconstruct stool from regional measurements by nested leave one participant out validation under complementary compositional geometries. We then assessed decision preservation through three spatial proxy states and measurement induced decision loss (MIDL) across normalized binary intervention thresholds. Multi region CLR ridge and Aitchison models combining regional measurements with a population reference had the lowest point estimate Aitchison errors, but evidence that either outperformed the population Aitchison center was not retained after multiplicity correction. Rankings changed under total variation and after guild amalgamation; at the coarser resolution, the primary ordering showed a small, uncertain reversal that was sensitive to zero handling. The model with the lowest compositional error did not have the lowest MIDL curve calibration error. Population centers closely reproduced the nearly uninformative MIDL curves of observed stool while retaining little individualized regional information. These findings show that reconstruction, decision preservation, and calibration to observed stool decision loss answer different questions. Evaluation of microbiome measurement strategies should therefore treat compositional geometry, biological representation, and the downstream decision task as parts of the estimand.

systems biology↗

bioRxiv2026

Molecular Force Gates Regulate Mechanical Transmission Across the Nuclear Envelope

The LINC (linker of nucleoskeleton and cytoskeleton) complex is a central mechanotransductive apparatus that physically couples cytoskeletal forces to the nuclear interior, thereby regulating nuclear architecture, chromatin organization, and downstream signaling. Using atomistic molecular dynamics simulations, we identify and characterize force gates within the LINC complex, structural interactions that maintain mechanical connectivity under load. By introducing targeted mutations into the KASH peptide and SUN protein, we provide computational evidence that a key disulfide bond (SUN2 Cys563 to KASH2 Cys6862) and specific residues within the KASH {beta} strand interaction motif (KASH6875, KASH6877) act as distinct mechanical elements during force transmission across the SUN and KASH interface. Leveraging this framework, we show that the disulfide bond and the {beta} sheet network contribute unequally: removing the disulfide reduces both sustained and peak force transmission approximately twofold, whereas disrupting the {beta} sheet network leaves peak force largely intact while altering interfacial coupling and the duration over which load is borne. Localized molecular changes therefore reshape mechanical coupling to the nucleus through distinct routes rather than through a single graded loss of force. Such altered LINC mechanics may be relevant to laminopathies, including Hutchinson Gilford Progeria Syndrome (HGPS), in which defective nuclear lamina compromises force transmission; however, implications for specific diseases remain speculative and require experimental validation. By resolving how distinct molecular contacts contribute to force transmission in space and sequence, this work establishes a framework for dissecting mechanical coupling at protein interfaces under load. These findings provide a foundation for future studies aimed at experimentally testing how individual interface interactions govern nuclear force transmission, a prerequisite for therapeutically targeting LINC mechanics. More broadly, the force gate framework may guide strategies to modulate interfaces that bear mechanical loads in other mechanically regulated proteins. Together, these findings establish a computational framework for investigating force transmission at mechanically loaded protein interfaces and provide experimentally testable predictions for future studies.

cell biology↗

bioRxiv2026

Measurement-Induced Decision Loss: A feasible-set theory of microbiome intervention under observation loss

Microbiome measurements are commonly judged by how faithfully they reconstruct an underlying biological profile, using taxonomic similarity, prediction accuracy, or compositional distance. Yet faithfulness and usefulness are not the same: a distorted measurement may still support the correct intervention, while a small error can be decisive when it falls near the boundary between two treatment choices. We therefore evaluate measurement quality by whether information loss changes the decisions and outcomes that remain attainable. We treat the intervention-relevant state of the gut community as a hidden quantity seen only through an imperfect, information losing measurement, mapped by a decision rule to a graded intervention. Each decision rule is evaluated by net host benefit and ecological margin, generating a feasible set of attainable outcome pairs. We define the measurement-induced decision loss (MIDL) as the maximum net host benefit lost under imperfect measurement relative to perfect observation. For scalar affine information loss, increasing degradation is a Blackwell garbling; under post-processing closure, more informative measurement weakly expands the feasible set and cannot reduce optimal host benefit. In the binary-state model, MIDL increases linearly while an observation-dependent decision rule remains optimal and plateaus once the optimal policy becomes signal-independent. We extend the framework to finite states, state-specific loss, multi-segment stool aggregation, and ecological-margin constraints. The framework distinguishes measurement faithfulness from decision usefulness and provides a principled basis for comparing stool and richer measurements by the interventions and trade-offs they support.

systems biology↗

bioRxiv2026

Identifiability of metabolic resilience from sparse longitudinal metabolomics

Sparse, irregular longitudinal metabolomic sampling fundamentally constrains which dynamical properties of gut metabolism can be robustly inferred from observational data. We develop an effective landscape inference framework to characterize these identifiability limits while quantifying aspects of metabolic resilience that remain recoverable under realistic sampling regimes. Using stochastic simulations with known ground truth, we first characterize the identifiability limits of multistability detection under sparse sampling, showing that bistable dynamics can appear monostable at sampling densities typical of existing human cohorts. Guided by these identifiability limits, we illustrate the framework using a small subset (N = 4) of longitudinal stool metabolomic trajectories meeting stringent quality-control criteria. Within this sparse-sampling regime, landscape curvature, a bootstrap-quantified measure of local recovery strength, remains identifiable and provides preliminary evidence of inter-individual variability in inferred recovery dynamics. An autoregressive extension prioritizes bile acids and fermentation intermediates as candidate modulators of butyrate return dynamics.

microbiology↗

bioRxiv2025

Semi-supervised Retrieval of Functional Residues Through the Integration of Protein Language Models and Gene Ontology Data

MotivationExperimental studies of protein function often focus on mechanistic descriptions, characterizing how specific sites and residues contribute to activity. Abstractions such as domains and active sites enable quantitative descriptions of how protein features act biologically. Thanks to the abundance of high-quality sequence and function data, machine learning has achieved great success in directly predicting protein function. However, translating functional characterizations into mechanistic ones on the level of the domains, binding sites, or motifs remains challenging. This represents a semi-supervised problem: sequences and global functional labels are available, but local annotations must be inferred. ResultsWe investigate the unsupervised discovery of functionally active protein regions by integrating protein sequence models with functional information. We first formalize the residue-level functional annotation problem by constructing unified evaluation datasets linking Gene Ontology functions to annotated residues. Eight datasets are assembled, spanning levels of specificity from single active-site residues to domains covering up to 60% of a protein. We then introduce a new class of function-conditioned generative models that more accurately predict functionally important residues than existing approaches, including interpretability methods and PSSM entropy estimation, across multiple benchmark datasets. Availabilitygithub.com/mofradlab/go_interp Contactmofrad@berkeley.edu

bioinformatics↗

bioRxiv2025

KODA: Agentic Framework for Microbiome Drug Target Discovery

The gut microbiome plays a crucial role in human health and disease, influencing diverse biological processes such as immune regulation and nutrient metabolism. However, the complexity of micro-bial interactions and their metabolic cross-feeding dynamics remains poorly understood. This study proposes KODA, an agentic framework that integrates large language models (LLMs) and knowledge graphs (KGs) to facilitate the discovery of targets in antimicrobial drugs in the gut microbiome. Our approach employs a multi-agent system to interpret natural language queries and translate them into precise graph database queries, enabling intuitive interactions with complex microbiome data. Focusing on KEGG orthologies related to essential microbial genes, KODA identifies potential antimicrobial drug targets by analyzing microbial metabolic pathways. The system employs a Neo4j-based microbiome KG, which integrates microbial interaction data, metabolic models, and KEGG annotations. A dedicated evaluation framework, which incorporates LLM-based reviewers, assesses the quality of generated queries and analytical reports. Our results demonstrate the efficacy of KODA in providing actionable insights for antimicrobial research, particularly in identifying conserved essential genes as potential drug targets. This framework holds the potential to democratize microbiome research by lowering technical barriers and accelerating hypothesis generation in drug discovery.

bioinformatics↗

bioRxiv2025

SIMBA-GNN: Simulation-augmented Microbiome Abundance Graph Neural Network

Understanding gut microbiome dynamics gut requires deciphering complex, metabolically driven interactions beyond taxonomic profiles. We present SIMBA, a novel framework that integrates mechanistic metabolic simulations with a graph neural network (GNN) to predict microbial abundances and uncover cross-feeding relationships. By simulating pairwise interactions among gut microbes using metabolic networks, we generate biologically grounded graphs that capture metabolite cross-feeding and functional relationships. Our custom GNN, enhanced with edge-aware attention, is trained through a multi-stage pipeline combining self-supervised learning, simulation-based pretraining, and fine-tuning on real microbial abundance data. SIMBA achieves state-of-the-art performance (Spearman {rho} = 0.85) and enables interpretable insights into keystone taxa and metabolic bottlenecks. This work demonstrates the power of combining metabolic networks with deep learning for precision microbiome analysis.

systems biology↗

bioRxiv2024

MetaBiome: A Multiscale Model Integrating Agent-Based Modeling and Metabolic Networks Reveals Spatial Regulation in Mucosal Microbial Communities

Mucosal microbial communities (MMCs) are complex ecosystems near the mucosal layers of the gut, essential for maintaining health and modulating disease states. Despite advances in high-throughput omics technologies, current methodologies struggle to capture the dynamic metabolic interactions and spatiotemporal variations within MMCs. In this work, we present MetaBiome, a multiscale model integrating agent-based modeling (ABM), finite volume methods, and constraint-based models to explore the metabolic interactions within these communities. Integrating ABM allows for the detailed representation of individual microbial agents, each governed by rules that dictate cell growth, division, and interactions with their surroundings. Through a layered approach--encompassing environmental conditions, agent information, and metabolic pathways--we simulated different communities to showcase the potential of the model. Using our in-silico platform, we explored the dynamics and spatiotemporal patterns of MMCs in the proximal small intestine and the cecum, simulating the physiological conditions of the two gut regions. Our findings revealed how specific microbes adapt their metabolic processes based on substrate availability and local environmental conditions, shedding light on spatial metabolite regulation and informing targeted therapies for localized gut diseases. MetaBiome provides a detailed representation of microbial agents and their interactions, surpassing the limitations of traditional grid-based systems. This work marks a significant advancement in microbial ecology as it offers new insights into predicting and analyzing microbial communities. ImportanceOur study presents a novel multiscale model that combines agent-based modeling, finite volume methods, and genome-scale metabolic models to simulate the complex dynamics of mucosal microbial communities in the gut. This integrated approach allows us to capture spatial and temporal variations in microbial interactions and metabolism that are difficult to study experimentally. Key findings from our model include: O_LIPrediction of metabolic cross-feeding and spatial organization in multi-species communities C_LIO_LIInsights into how oxygen gradients and nutrient availability shape community composition in different gut regions C_LIO_LIIdentification of spatially-regulated metabolic pathways and enzymes in E. coli C_LI We believe this work represents a significant advance in computational modeling of microbial communities and provides new insights into the spatial regulation of gut microbiome metabolism. The multiscale modeling approach we have developed could be broadly applicable for studying other complex microbial ecosystems.

microbiology↗

bioRxiv2024

Expanding the HP1a-binding consensus and molecular grammar for heterochromatin assembly

The recruitment of Heterochromatin Protein 1 (HP1) partners is essential for heterochromatin assembly and function, yet our knowledge regarding their organization in heterochromatin remains limited. Here we show that interactors engage the Drosophila HP1 (HP1a) dimer through a degenerate and expanded form of the previously identified PxVxL motif, which we now term HP1a Access Codes (HACs). These HACs reside in disordered regions, possess high conservation among Drosophila homologs, and contain alternating hydrophobic residues nested in a cluster of positively charged amino acids. These findings and molecular dynamics simulations identify key electrostatic interactions that modulate HP1a-binding strength and provide a dramatically improved HP1a-binding consensus motif that can reveal protein partners and the molecular grammar involved in heterochromatin assembly. We propose HP1a acts as a scaffold for other heterochromatin components containing HAC motifs, which in turn may regulate the function and higher order structure of the heterochromatin compartment.

molecular biology↗

bioRxiv2024

A Suite of Foundation Models Captures the Contextual Interplay Between Codons

In the canonical genetic code, many amino acids are assigned more than one codon. Work by us and others has shown that the choice of these synonymous codon is not random, and carries regulatory and functional consequences. Existing protein foundation models ignore this context-dependent role of coding sequence in shaping the protein landscape of the cell. To address this gap, we introduce cdsFM, a suite of codon-resolution large language models, including both EnCodon and DeCodon models, with up to 1B parameters. Pre-trained on 60 million protein-coding sequences from more than 5,000 species, our models effectively learn the relationship between codons and amino acids, recapitualing the overall structure of the genetic code. In addition to outperforming state-of-the-art genomic foundation models in a variety of zero-shot and few-shot learning tasks, the larger pre-trained models were superior in predicting the choice of synonymous codons. To systematically assess the impact of synonymous codon choices on protein expression and our models ability to capture these effects, we generated a large dataset measuring overall and surface expression levels of three proteins as a function of changes in their synonymous codons. We showed that our EnCodon models could be readily fine-tuned to predict the contextual consequences of synonymous codon choices. Armed with this knowledge, we applied EnCodon to existing clinical datasets of synonymous variants, and we identified a large number of synonymous codons that are likely pathogenic, several of which we experimentally confirmed in a cellbased model. Together, our findings establish the cdsFM suite as a powerful tool for decoding the complex functional grammar underlying the choice of synonymous codons.

genomics↗

bioRxiv2023

Integrin Mechanosensing relies on Pivot-clip Mechanism to Reinforce Cell Adhesion

Cells intricately sense mechanical forces from their surroundings, driving biophysical and biochemical activities. This mechanosensing phenomenon occurs at the cell-matrix interface, where mechanical forces resulting from cellular motion, such as migration or matrix stretching, are exchanged through surface receptors, primarily integrins, and their corresponding matrix ligands. A pivotal player in this interaction is the 5{beta}1 integrin and fibronectin (FN) bond, known for its role in establishing cell adhesion sites for migration. However, upregulation of the 5{beta}1-FN bond is associated with uncontrolled cell metastasis. This bond operates through catch bond dynamics, wherein the bond lifetime paradoxically increases with greater force. The mechanism sustaining the characteristic catch bond dynamics of 5{beta}1-FN remains unclear. Leveraging molecular dynamics simulations, our approach unveils a pivot-clip mechanism. Two key binding sites on FN, namely the synergy site and the RGD (arg-gly-asp) motif, act as active points for structural changes in 5{beta}1 integrin. Conformational adaptations at these sites are induced by a series of hydrogen bond formations and breaks at the synergy site. We disrupt these adaptations through a double mutation on FN, known to reduce cell adhesion. A whole-cell finite element model is employed to elucidate how the synergy site may promote dynamic 5{beta}1-FN binding, resisting cell contraction. In summary, our study integrates molecular and cellular-level modeling to propose that FNs synergy site reinforces cell adhesion through enhanced binding dynamics and a mechanosensitive pivot-clip mechanism. This work sheds light on the interplay between mechanical forces and cell-matrix interactions, contributing to our understanding of cellular behaviors in physiological and pathological contexts. SIGNIFICANCE5{beta}1 integrin serves as a mediator of cell-matrix adhesion and has garnered attention as a target for impeding cancer metastasis. Despite its importance, the mechanism underlying the formation of a catch bond between 5{beta}1 integrin and its primary ligand, fibronectin, has remained elusive. Our study aims to address this gap by proposing a pivot-clip mechanism. This mechanism elucidates how 5{beta}1 integrin and fibronectin collaboratively reinforce cell adhesion through conformational changes induced by the dynamic interaction of a key binding motif known as the synergy site.

biophysics↗

bioRxiv2023

Multiscale Computational Framework to Investigate Integrin Mechanosensing and Cell Adhesion

Integrin mechanosensing plays an instrumental role in cell behavior, phenotype, and fate by transmitting mechanical signals that trigger downstream molecular and cellular changes. For instance, force transfer along key amino acid residues can mediate cell adhesion. Disrupting key binding sites within 5{beta}1 integrins binding partner, fibronectin (FN) diminishes adhesive strength. While past studies have shown the importance of these residues in cell adhesion, the relationship between the dynamics of these residues and how integrin distributes force across the cell surface remains less explored. Here, we present a multiscale mechanical model to investigate the mechanical coupling between integrin nanoscale dynamics and whole-cell adhesion mechanics. Our framework leverages molecular dynamics simulations to investigate residues within 5{beta}1-FN during stretching and the finite element method to visualize the whole-cell adhesion mechanics. The forces per integrin across the cell surface of the whole-cell model were consistent with past atomic force microscopy and Forster resonance energy transfer measurements from literature. The molecular dynamics simulations also confirmed past studies that implicate two key sites within FN that maintain cell adhesion: the synergy site and RGD motif. Our study contributed to our understanding of molecular mechanisms by which these sites collaborate to mediate whole-cell integrin adhesion dynamics. Specifically, we showed how FN unfolding, residue binding/unbinding, and molecular structure contribute to 5{beta}1-FNs nonlinear force-extension behavior during stretching. Our computational framework could be used to explain how the dynamics of key residues influence cell differentiation or how uniquely designed protein structures could dynamically limit the spread of metastatic cells.

biophysics↗

bioRxiv2020

EpitopeVec: Linear Epitope Prediction Using DeepProtein Sequence Embeddings

MotivationB-cell epitopes (BCEs) play a pivotal role in the development of peptide vaccines, immunodiagnostic reagents, and antibody production, and thus generally in infectious disease prevention and diagnosis. Experimental methods used to determine BCEs are costly and time-consuming. It thus becomes essential to develop computational methods for the rapid identification of BCEs. Though several computational methods have been developed for this task, cross-testing of classifiers trained and tested on different datasets revealed their limitations, with accuracies of 51 to 53%. ResultsWe describe a new method called EpitopeVec, which utilizes residue properties, modified antigenicity scales, and a Protvec representation of peptides for linear BCE prediction with machine learning techniques. Evaluating on several large and small data sets, as well as cross-testing demonstrated an improvement of the state-of-the-art performances in terms of accuracy and AUC. Predictive performance depended on the type of antigen (viral, bacterial, eukaryote, etc.). In view of that, we also trained our method on a large viral dataset to create a linear viral BCE predictor. AvailablityThe software is available at https://github.com/hzi-bifo/epitope-prediction under the GPL3.0 license. Contactalice.mchardy@helmholtz-hzi.de Supplementary informationSupplementary data are available at Bioinformatics online.

bioinformatics↗

bioRxiv2019

DeepPrime2Sec: Deep Learning for Protein Secondary Structure Prediction from the Primary Sequences

MotivationHere we investigate deep learning-based prediction of protein secondary structure from the protein primary sequence. We study the function of different features in this task, including one-hot vectors, biophysical features, protein sequence embedding (ProtVec), deep contextualized embedding (known as ELMo), and the Position Specific Scoring Matrix (PSSM). In addition to the role of features, we evaluate various deep learning architectures including the following models/mechanisms and certain combinations: Bidirectional Long Short-Term Memory (BiLSTM), convolutional neural network (CNN), highway connections, attention mechanism, recurrent neural random fields, and gated multi-scale CNN. Our results suggest that PSSM concatenated to one-hot vectors are the most important features for the task of secondary structure prediction.\n\nResultsUtilizing the CNN-BiLSTM network, we achieved an accuracy of 69.9% and 70.4% using ensemble top-k models, for 8-class of protein secondary structure on the CB513 dataset, the most challenging dataset for protein secondary structure prediction. Through error analysis on the best performing model, we showed that the misclassification is significantly more common at positions that undergo secondary structure transitions, which is most likely due to the inaccurate assignments of the secondary structure at the boundary regions. Notably, when ignoring amino acids at secondary structure transitions in the evaluation, the accuracy increases to 90.3%. Furthermore, the best performing model mostly mistook similar structures for one another, indicating that the deep learning model inferred high-level information on the secondary structure.\n\nAvailabilityThe developed software called DeepPrime2Sec and the used datasets are available at http://llp.berkeley.edu/DeepPrime2Sec.\n\nContactmofrad@berkeley.edu

bioinformatics↗

bioRxiv2019

Fighting antimicrobial resistance in Pseudomonas aeruginosa with machine learning-enabled molecular diagnostics

The growing importance of antibiotic resistance on clinical outcomes and cost of care underscores the need for optimization of current diagnostics. For a number of bacterial species antimicrobial resistance can be unambiguously predicted based on their genome sequence. In this study, we sequenced the genomes and transcriptomes of 414 drug-resistant clinical Pseudomonas aeruginosa isolates. By training machine learning classifiers on information about the presence or absence of genes, their sequence variation, and gene expression profiles, we generated predictive models and identified biomarkers of susceptibility or resistance to four commonly administered antimicrobial drugs. Using these data types alone or in combination resulted in high (0.8-0.9) or very high (>0.9) sensitivity and predictive values, where the relative contribution of the different categories of biomarkers strongly depended on the antibiotic. For all drugs except for ciprofloxacin, gene expression information substantially improved diagnostic performance. Our results pave the way for the development of a molecular resistance profiling tool that reliably predicts antimicrobial susceptibility based on genomic and transcriptomic markers. The implementation of a molecular susceptibility test system in routine clinical microbiology diagnostics holds promise to provide earlier and more detailed information on antibiotic resistance profiles of bacterial pathogens and thus could change how physicians treat bacterial infections.

microbiology↗