Biology subjects

Sriram, A.

Publications and source records attributed to Sriram, A..

10 recordsLinked to original sources

bioRxiv2026

Prediction transfer does not imply reliability transfer in single-cell foundation models

Single-cell foundation models are increasingly transferred across tissues, diseases, perturbation screens, and experimental platforms. Although substantial effort has focused on prediction transfer, it remains unclear whether a model's assessment of its own reliability transfers similarly across biological contexts. Here we define reliability transfer as the ability to transfer an accurate assessment of when model predictions are likely to be correct or incorrect and investigate its behavior across ten perturbation-prediction systems spanning multiple datasets, models, and evaluation settings. We find that reliability transfer varies substantially across source-target pairs and cannot be inferred from predictive performance alone, indicating that prediction transfer and reliability transfer constitute distinct forms of generalization. Analyses of biological annotations, network connectivity, target exposure, and contextual factors reveal that prediction reliability is associated with both shared and context-dependent signals, with their relative importance differing across systems. To characterize this variability, we introduce reliability-attribution fingerprints, context-level signatures derived from strictly cross-fitted SHAP explanations that summarize how reliability estimators combine biological and contextual information when assigning confidence. Across systems, contexts with similar attribution-magnitude fingerprints tended to transfer reliability more successfully, and fingerprint similarity identified transfer sources achieving higher reliability-transfer performance within the observed candidate set. However, these relationships weakened under stricter cross-family transfer settings, indicating that transferable reliability structure remains partly dependent on shared contextual information. Finally, transferred reliability estimators produced different perturbation rankings despite identical perturbation predictions, illustrating a practical consequence of reliability transfer for downstream prioritization. Together, these results establish reliability transfer as a distinct aspect of foundation-model generalization and introduce reliability-attribution fingerprints as a framework for characterizing transferable reliability structure across biological contexts.

bioinformatics↗

bioRxiv2026

BUDDY: An interactive web platform to accelerate the functional modeling of human missense variants using Saccharomyces cerevisiae orthologs

The advent of next-generation sequencing has transformed clinical genetics, accelerating the identification of human genetic variants. However, the labor-intensive process of functional validation of Variants of Uncertain Significance (VUS) remains a significant diagnostic barrier. While expressing humanized genes in Saccharomyces cerevisiae is a common approach for modeling missense variants with functional assays, using native yeast orthologs provides a more streamlined approach that preserves species-specific interactions within multisubunit complexes. However, taking this approach relies on a fragmented workflow that requires bioinformatic experience. To address this issue, we developed the Bioinformatic Utility for Diagnostic Discovery in Yeast (BUDDY), an integrated web platform designed to streamline variant analysis by connecting bioinformatic tools and databases including ClinVar. Features include amino acid alignments, a protein structure viewer that displays a superposition of yeast and human proteins, pathogenicity prediction data, CRISPR-Cas9 reagent design, and yeast phenotypes to provide potential functional assays. The platform also contains an artificial intelligence (AI)-assisted experimental design feature that evaluates the utility of modeling specific variant and gene combinations in yeast and can assist with phenotype selection. BUDDY significantly accelerated the bioinformatic preparation time needed for ortholog variant modeling by an experienced biologist. Additionally, without prior bioinformatics training, a first-year medical student used BUDDY to develop a self-directed project and proposed a study of five Lynch syndrome-associated VUS using a forward mutation rate assay. BUDDY is designed to accelerate the investigation of human variants in yeast and has a high potential utility for use in resource-constrained settings such as course-based undergraduate research experiences (CUREs).

bioinformatics↗

bioRxiv2026

Airborne Sound Detection in Manduca sexta Caterpillars

We investigated the hearing mechanism of Manduca sexta caterpillars to determine whether their behavioural responses to acoustic stimulation are caused by sound-induced substrate vibration or by direct detection of airborne sound. Behavioural response thresholds to direct substrate vibration and airborne sound were measured using pure-tone stimuli at 150 Hz and 2000 Hz. A frequency of 150 Hz was selected based on previous behavioural measurements showing the lowest sound detection thresholds between approximately 100 and 200 Hz, while 2000 Hz was selected based on laser Doppler vibrometry measurements that revealed a mechanical resonance near this frequency in a thoracic hair. At both test frequencies, the direct substrate vibration detection thresholds were more than 100 times greater, on average, than the sound-induced substrate vibrations measured at the behavioural sound detection thresholds. Thus, the behavioural responses to airborne sound cannot be explained by sound-induced substrate vibration. During airborne sound stimulation, caterpillars exhibited graded behavioural responses with increasing sound pressure level, progressing from freezing to twitching and finally to jump-startle. Together, these findings provide direct experimental evidence that the behavioural responses of M. sexta caterpillars to acoustic stimulation cannot be explained by sound-induced substrate vibration and instead support direct detection of airborne sound. Understanding how caterpillars respond to airborne sound may also provide biological inspiration for the development of bio-inspired acoustic sensors.

animal behavior and cognition↗

bioRxiv2026

eQTM (expression quantitative trait methylation) Atlas: a comprehensive resource of over 11 million DNA methylation-gene expression associations through across 11 tissues and 4 diseases

MotivationEpigenome-wide association studies (EWAS) have identified numerous DNA methylation (DNAm) CpG sites associated with complex traits and diseases, but interpretation of those CpG sites remains challenging because in EWAS, CpGs are mostly linked to nearby genes based only on genomic proximity. Expression quantitative trait methylation (eQTM) analyses connect DNAm CpGs with statistically associated gene expression levels. However, a comprehensive, searchable resource integrating eQTMs across diverse tissues and disease contexts has been lacking. ResultsWe developed the eQTM Atlas, a web-based resource that manually curates more than 11 million DNAm-gene expression associations from eight cohorts, covering 11 tissue types, four broad disease contexts, 173,886 unique CpG probes and 20,231 unique genes. The Atlas supports gene- or CpG-searches by tissue or disease type and finding associated CpG or genes, visualization of cis- and trans-eQTMs through genome browser, heatmap interfaces across various tissues, and cohort-level data downloads. By integrating eQTM results with EWAS resources, the eQTM Atlas enables users to connect disease- or trait-associated CpGs to statistically associated genes rather than relying solely on proximity-based gene annotation, supporting functional interpretation of EWAS findings and generation of disease-specific regulatory hypotheses. Availability and implementationThe eQTM Atlas is freely available at https://shiny.crc.pitt.edu/eqtm_browser/. The web interface is implemented in R Shiny and hosted through the University of Pittsburgh Center for Research Computing (CRC). Source code is available at https://github.com/ads303/eQTM-Atlas.

genetics↗

bioRxiv2026

Asexual Adaptation Drives Transient and Reversible Changes in Mating Efficiency.

The evolution of reproductive isolation is central to speciation, yet the earliest stages of this process remain poorly understood. In particular, it is unclear how rapidly barriers to mating arise during adaptation, whether they accumulate predictably, and how they depend on ecological context. Here, we investigate the evolution of mating efficiency during prolonged asexual adaptation in diploid Saccharomyces cerevisiae. Twelve replicate populations were evolved for 1200 generations in two distinct carbon environments, glucose and galactose, under strictly asexual conditions. At regular intervals, we induced sporulation and quantified mating efficiency using three complementary assays: within-population crosses, crosses between populations evolved in different environments, and crosses between evolved populations and the ancestral strain. We find that mating efficiency evolves during asexual adaptation, with outcomes that depend strongly on the environment. While glucose-evolved populations remain largely stable, galactose-evolved populations exhibit a reversible decline. Overall, changes in mating efficiency are dynamic, heterogeneous, and often transient, with evidence for both intrinsic reductions in mating competence and context-dependent incompatibilities between populations. Together, these results show that asexual adaptation can generate rapid but non-monotonic changes in mating compatibility. Early reductions in mating efficiency are heterogeneous, environment-dependent, and often reversible, and do not accumulate into stable reproductive isolation over the timescale examined. Our findings suggest that the initial stages of divergence are characterized by dynamic and contingent perturbations of reproductive traits, rather than a steady progression toward speciation.

evolutionary biology↗

bioRxiv2025

DeepDiff-SHAP: Interpretable deep learning for subgroup-specific causal inference using conditional SHAP

Precision medicine aims to tailor healthcare strategies to individual differences in genetic, clinical, and environmental factors. However, identifying subgroup-specific causal relationships in complex biomedical data remains a major challenge, especially when standard causal inference methods average over population heterogeneity. We introduce DeepDiff-SHAP, a novel framework that combines regression-based and deep learning-based differential causal inference to detect changes in causal relationships across patient subgroups. DeepDiff-SHAP integrates SHapley Additive exPlanations (SHAP) to estimate conditional dependencies and perform nonlinear differential causal inference in a principled, interpretable manner. Applying DeepDiff-SHAP to two population-scale datasets, the CDC Diabetes Health Indicators Dataset and a UK Biobank sepsis cohort stratified by hypertension, we identified clinically meaningful and subgroup-specific causal changes in relationships between features such as age, general health, alkaline phosphatase, and cholesterol. Our results demonstrate that deep learning enhances sensitivity to complex interaction patterns overlooked by linear models, providing new biological insights into disease progression and comorbidity-specific risk mechanisms. DeepDiff-SHAP offers a scalable and interpretable solution to uncover individualized causal pathways, advancing the goal of truly personalized medicine.

bioinformatics↗

bioRxiv2025

Episodic pain in Fabry disease is mediated by a heat shock protein-TRPA1 axis.

Two-thirds of patients with Fabry disease suffer debilitating pain attacks triggered by exercise, fever, and exposure to environmental heat. These patients face an even greater risk of heat-related episodic pain in the face of global climate change. Almost nothing is known about the biological mechanisms underlying heat-induced pain crises in Fabry disease, and there is no preclinical model available to study Fabry crises. Here, we established the first model of heat-induced pain attacks in Fabry disease by exposing transgenic Fabry rats to environmental heat. Heat exposure precipitated robust mechanical hypersensitivity, closely matching temporal features reported by patients with Fabry disease. At the cellular level, heat exposure sensitized Fabry dorsal root ganglia (DRG) neurons to agonists for transient receptor potential cation channel A1 (TRPA1), but not TRPV1. The heat shock response, which normally confers heat-resilience, was impaired in Fabry disease, and we demonstrated that heat shock proteins (HSP70 and HSP90) regulate TRPA1. Strikingly, pharmacologically inhibiting HSP90 completely prevented cellular and behavioral sensitization by environmental heat in Fabry disease. Together, this work establishes the first model of episodic pain in Fabry disease, implicates the heat shock response in heat-evoked pain episodes, and identifies a novel heat shock protein-TRPA1 regulatory axis. O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=96 SRC="FIGDIR/small/639340v2_ufig1.gif" ALT="Figure 1"> View larger version (23K): org.highwire.dtl.DTLVardef@14e655eorg.highwire.dtl.DTLVardef@1e13455org.highwire.dtl.DTLVardef@32c5b9org.highwire.dtl.DTLVardef@152fcfd_HPS_FORMAT_FIGEXP M_FIG C_FIG

neuroscience↗

bioRxiv2024

Unveiling the Gut Microbiome Network Fingerprint: A Deep Learning Approach to Predict Immunotherapy Response in Melanoma Patients.

Immune checkpoint inhibitors (ICI) are critical in cancer therapy, harnessing the immune system to fight tumors. The gut microbiome, a diverse ecosystem of trillions of microorganisms, has emerged as a key influencer of ICI efficacy. However, the specific gut bacteria linked to ICI response in melanoma patients remain unclear, with studies showing inconsistent findings across cohorts. Attributing the lack of consensus to multiple layers of non-linear interactions with host factors and within the microbial community, we developed DeepMicroNET, a novel deep neural network (DNN) model-driven network analysis method to identify and characterize the gut bacteria associated with treatment response. Utilizing data from five cohorts of advanced melanoma patients undergoing ICI therapies, we demonstrate how DeepMicroNET identifies gut bacteria consistently predicting immunotherapy responses (response-associated or RA bacteria). Further, DeepMicroNET revealed that these bacteria are associated with key immune pathways, such as antigen processing and presentation, offering novel insights into the molecular mechanisms that could be targeted to improve therapeutic outcomes. Altogether, by identifying gut bacteria robustly associated with immunotherapy response, DeepMicroNET opens the door to a new era of personalized medicine based on the gut microbiome profile.

microbiology↗

bioRxiv2023

Keratinocyte Piezo1 drives paclitaxel-induced mechanical hypersensitivity

Recent work demonstrates that epidermal keratinocytes are critical for normal touch sensation. However, it is unknown if keratinocytes contribute to touch evoked pain and hypersensitivity following tissue injury. Here, we used inhibitory optogenetic and chemogenetic techniques to determine the extent to which keratinocyte activity contributes to the severe neuropathic pain that accompanies chemotherapeutic treatment. We found that keratinocyte inhibition largely alleviates paclitaxel-induced mechanical hypersensitivity. Furthermore, we found that paclitaxel exposure sensitizes mouse and human keratinocytes to mechanical stimulation through the keratinocyte mechanotransducer Piezo1. These findings demonstrate the contribution of non-neuronal cutaneous cells to neuropathic pain and pave the way for the development of new pain-relief strategies that target epidermal keratinocytes and Piezo1. SummarySensitization of the keratinocyte mechanotransducer Piezo1 drives paclitaxel-induced touch pain.

neuroscience↗

bioRxiv2020

Assessment of autism zebrafish mutant models using a high-throughput larval phenotyping platform

In recent years zebrafish have become commonly used as a model for studying human traits and disorders. Their small size, high fecundity, and rapid development allow for more high-throughput experiments compared to other vertebrate models. Given that zebrafish share >70% gene homologs with humans and their genomes can be readily edited using highly efficient CRISPR methods, we are now able to rapidly generate mutations impacting practically any gene of interest. Unfortunately, our ability to phenotype mutant larvae has not kept pace. To address this challenge, we have developed a protocol that obtains multiple phenotypic measurements from individual zebrafish larvae in an automated and parallel fashion, including morphological features (i.e., body length, eye area, and head size) and movement/behavior. By assaying wild-type zebrafish in a variety of conditions, we determined optimal parameters that avoid significant developmental defects or physical damage; these include morphological imaging of larvae at two time points (3 days post fertilization (dpf) and 5 dpf) coupled with motion tracking of behavior at 5 dpf. As a proof-of-principle, we tested our approach on two novel CRISPR-generated mutant zebrafish lines carrying predicted null-alleles of syngap1b and slc7a5, orthologs to two human genes implicated in autism-spectrum disorder, intellectual disability, and epilepsy. Using our optimized high-throughput phenotyping protocol, we recapitulated previously published results from mouse and zebrafish models of these candidate genes. In summary, we describe a rapid parallel pipeline to characterize morphological and behavioral features of individual larvae in a robust and consistent fashion, thereby improving our ability to better identify genes important in human traits and disorders. AUTHOR SUMMARYZebrafish (Danio rerio) are a well-established model organism for the study of neurodevelopmental disorders. Due to their small size, fast reproduction, and genetic homology with humans, zebrafish have been widely used for characterizing and screening candidate genes for many disorders, including autism-spectrum disorder, intellectual disability, and epilepsy. Although several studies have described the use of high-throughput morphological and behavioral assays, few combine multiple assays in a single zebrafish larva. Here, we optimized a platform to characterize morphometric features at two developmental time points in addition to behavioral traits of zebrafish larvae. We then used this approach to characterize two autism candidate genes (SYNGAP1 and SLC7A5) in two CRISPR-generated zebrafish null mutant models we developed in house. These data recapitulate previously published results related to enhanced seizure activity, while identifying additional defects not previously reported. We propose that our phenotyping platform represents a feasible method for maximizing the use of single zebrafish larvae in the characterization of additional mutants relevant to neurodevelopmental disorders.

genetics↗