Biology subjects

Sanghi, S.

Publications and source records attributed to Sanghi, S..

6 recordsLinked to original sources

bioRxiv2025

SigSpace: an LLM-based agent for drug responsesignature interpretation

Agent systems powered by large language models (LLMs) are increasingly applied in computational biology to automate analysis, integrate data, and accelerate discovery. Here, we investigate the capacity of LLM-driven agents to interpret transcriptional response signatures of drug perturbations in cancer cell lines, a task central to understanding drug mechanisms of action (MoAs) and supporting cancer drug discovery. Leveraging the Tahoe-100M dataset of 100 million transcriptomic profiles across 1,100 small-molecule perturbations and 50 cancer cell lines, we developed an LLM-based agent system, SigSpace, that processes differential gene expression signatures and generates concise, human-readable summaries of drug responses. We then tested whether blinded response signature summaries could be correctly matched to their corresponding drug identity or MoA. Our results show that LLM-generated summaries consistently outperform random baselines and that the choice of LLM model and signature score significantly influences performance. These findings highlight the potential of LLMs to enhance interpretation of complex transcriptional data to enable drug discovery. Future directions for improvement include exploring alternative response signature formats, improving summarization fidelity, benchmarking performance across different summary formats and lengths, and broadening applications to additional datasets and predictive tasks in drug discovery.

bioinformatics↗

bioRxiv2025

Predicting emergent phenotypes from single cell populations using CELLECTION

Biological systems exhibit emergent phenotypes that arise from the collective behavior of individual components, such as whole-organ functions that arise from the coordinated activity of its individual cells, or organism-level phenotypes that result from the functional interplay of collections of genes in the genome. We present CELLECTION, a deep learning framework that learns to associate subgroups of instances with different emergent phenotypes. We show CELLECTION enables interpretable predictions for heterogeneous tasks, including disease classification, identification of disease-associated cell subtypes, alignment of developmental stages between human model systems, and even predicting relative hand-wing indices across the avian lineage. CELLECTION therefore provides a scalable and flexible framework for identifying key cellular or genetic signatures underlying complex traits in development, disease, and evolution.

genomics↗

bioRxiv2024

Oncogenic STIL-mediated loss of BRCA1 functionality causes DNA damage and centrosome amplification

DNA damages increase centrosome number, a typical phenotype in many cancers. However, the molecular linkers between the two pathways remain elusive. STIL (SCL/TAL1 Interrupting locus) is a core centrosome-organizing protein overexpressed in different cancer types. This work shows that STIL forms a complex with the DNA damage response protein, BRCA1, and is required for regulating its protein stability. Subsequently, STIL depletion resulting in BRCA1 loss from the nucleus causes centrosome amplification and DNA damage. We also identified a heterozygous missense mutation (S76L) in STIL from the pan-cancer datasets, which does not affect total BRCA1 protein levels. However, it reduces BRCA1 nuclear localization with enhanced centrosome recruitment simultaneously at the S-phase. The enhanced BRCA1 at centrosome causes an increase in Aurora-A kinase levels at centrosomes, thus culminating in DNA damage and centrosome amplification. The expression of exogenous BRCA1 can push back this imbalance of the STIL-BRCA1 axis in oncogenic conditions. The amplified centrosomes in oncogenic STIL condition cluster in the M phase and cause pseudo-bipolar spindle organization. We could decluster them by the HSET(KIFC1)-specific inhibitor, CW069, suggesting a promising target for anti-cancer treatment.

cell biology↗

bioRxiv2024

RNA3DB: a dataset for training and benchmarking deep learning models for RNA structure prediction

With advances in protein structure prediction thanks to deep learning models like AlphaFold, RNA structure prediction has recently received increased attention from deep learning researchers. RNAs introduce substantial challenges due to the sparser availability and lower structural diversity of the experimentally resolved RNA structures in comparison to protein structures. These challenges are often poorly addressed by the existing literature, many of which report inflated performance due to using training and testing sets with significant structural overlap. Further, the most recent Critical Assessment of Structure Prediction (CASP15) has shown that deep learning models for RNA structure are currently outperformed by traditional methods. In this paper we present RNA3DB, a dataset of structured RNAs, derived from the Protein Data Bank (PDB), that is designed for training and benchmarking deep learning models. The RNA3DB method arranges the RNA 3D chains into distinct groups (Components) that are non-redundant both with regard to sequence as well as structure, providing a robust way of dividing training, validation, and testing sets. Any split of these structurally-dissimilar Components are guaranteed to produce test and validations sets that are distinct by sequence and structure from those in the training set. We provide the RNA3DB dataset, a particular train/test split of the RNA3DB Components (in an approximate 70/30 ratio) that will be updated periodically. We also provide the RNA3DB methodology along with the source-code, with the goal of creating a reproducible and customizable tool for producing structurally-dissimilar dataset splits for structural RNAs. Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=140 SRC="FIGDIR/small/578025v3_ufig1.gif" ALT="Figure 1"> View larger version (39K): org.highwire.dtl.DTLVardef@cb3aa2org.highwire.dtl.DTLVardef@7685fdorg.highwire.dtl.DTLVardef@156ca35org.highwire.dtl.DTLVardef@f58d72_HPS_FORMAT_FIGEXP M_FIG C_FIG HighlightsO_LIWhile there is a recent surge in applying deep learning to RNA structure prediction, domain experts have raised concerns about generalization and current trends in benchmarking. C_LIO_LIMany of the concerns primarily relate to how novel RNA families-i.e. families unseen in the training set-are benchmarked, and whether the models are effective at handling such cases. Performance on bench-marks reflective of real-world applications, such as CASP15 and RNA-Puzzles, is poor for RNA deep learning models. C_LIO_LIWe present a dataset-RNA3DB-that is designed for training and bench-marking deep learning models for RNA structure prediction. RNA3DB provides coverage of all RNA chains found in the Protein Data Bank (PDB). C_LIO_LIRNA3DB is clustered into groups that are both sequentially and structurally non-redundant, providing a robust way of creating training, validation, and testing sets for deep learning models. Along with the dataset, we also provide a transparent methodology as well as the source-code, making our tool both reproducible and customizable. C_LI

bioinformatics↗

bioRxiv2023

Separation of function mutations in microcephaly protein CPAP/CENPJ, reveal its role in regulating centriole number and length.

Centriole are microtubule-based cylindrical structures characterized by their definite size, and stable, slow growing microtubules. The centriole core protein CPAP/CENPJ is known to act as a molecular cap regulating centriole length by interacting with microtubule/tubulin via the conserved microtubule destabilizing, PN2-3, and microtubule stabilizing, A5N, domains. The C-terminus of CPAP has a conserved glycine-rich G-box/TCP domain (1050-1338 amino acids). This region is involved in centriole cartwheel assembly by interacting with the cartwheel protein STIL. However, previously reported primary microcephaly mutation mapped in the G-box of CPAP, i.e., E1235V (MCPH6) affects centriole length via an unknown mechanism. Recently, another primary microcephaly mutation has been mapped to this region of CPAP, i.e., D1196N. However, the effect of D1196N on CPAP functioning is not known. We simultaneously characterized these two MCPH mutations in the G-box of CPAP. We identified that despite affecting the same domain of CPAP, they affect distinct CPAP functions at the centriole. The E1235V mutation caused an overly long centriole, and the D1196N mutation increased the centriole number. Interestingly, both these mutations affect CPAP direct interaction with the cartwheel protein STIL, which is involved in CPAP recruitment to the centriole. Accordingly, the CPAP E1235V centriole localization is significantly affected at the centriole. However, CPAP D1196N can still localize to centriole at levels comparable to the wild-type CPAP. We show that CPAP utilizes an alternate CEP152-dependent route for centriole recruitment. Importantly, our work highlights the importance of the CPAP region outside direct microtubule/tubulin interacting domains in influencing CPAP activity in cartwheel assembly and centriole length. Perhaps, this is why deleterious naturally occurring missense mutations are frequently occurring in this particular region of CPAP in primary microcephaly.

cell biology↗

bioRxiv2022

Activity and state dependent modulation of salt taste behavior in Drosophila melanogaster

Sodium present in NaCl is a fundamental nutrient required for many physiological processes. In animals, including Drosophila, low-salt concentrations induce attraction and high-salt concentrations evoke aversive behavior. Although the cellular basis of low and high salt taste detection has been described, the mechanisms regulating high salt consumption and salt taste modulation in animals remain elusive. Thus, we investigated the neural mechanisms behind high NaCl consumption in adult Drosophila, focusing on how flies adjust their acceptance of high salt based on their diet. Our findings show that long-term exposure to high salt increases the taste sensitivity of pharyngeal LSO neurons, which, in turn, enhances high salt intake. Exposing flies to a high NaCl diet for three days shows a decline in high salt aversion under starvation. Our results suggest that this modulation requires active LSO neurons and a starvation state or dopamine as genetic suppression of LSO pharyngeal neuronal activity in high NaCl-fed flies inhibit excessive salt intake. We also found that several independent taste receptor neurons and pathways are involved in such a modulation. Silencing any one of multiple LSO neuronal types inhibits excessive salt intake. Our study suggests that flies can adapt to the amount of salt ingested over several days, indicating the presence of a critical mechanism that resets the salt appetite and related neural circuits.

neuroscience↗