Biology subjects

De, M.

Publications and source records attributed to De, M..

7 recordsLinked to original sources

bioRxiv2026

Protein Language Model-Conditioned Graph Neural Networks for Multitask GPCR Ligand Activity Prediction

Predicting ligand activity across G protein-coupled receptors (GPCRs) requires models that capture both molecular structure and receptor-specific information while remaining robust to chemical and target-domain shifts. We developed a multimodal graph neural network that combines explicit ligand molecular graphs with frozen protein language model representations of GPCR sequences and jointly predicts quantitative pActivity and binary activity. The model was trained on 271,739 curated ligand-GPCR pairs spanning 183,694 ligands and 216 human GPCRs and evaluated using random, Bemis-Murcko scaffold, and strict cold-ligand partitions. With ESM-2 650M receptor embeddings, the selected model achieved mean absolute errors of 0.513 {+/-} 0.006, 0.540 {+/-} 0.006, and 0.641 {+/-} 0.005 pActivity units under random, cold-ligand, and scaffold evaluation, respectively, substantially outperforming a protein-aware fixed-feature multilayer perceptron and consistently improving quantitative prediction over a matched GINE reference. Removing receptor embeddings markedly degraded both regression and classification, whereas differences among ESM-2 35M, ESM-2 650M, and ProtT5 were comparatively small. Independent evaluation on 6,319 ChEMBL 37/BindingDB pairs revealed a substantial external domain shift, with mean absolute error increasing to approximately 0.94-0.95 despite chemically stringent internal validation. Nevertheless, receptor-specific information remained highly informative in a DRD2-DRD3 selectivity analysis: the cold-ligand ensemble achieved R^2 = 0.853, Spearman {rho} = 0.916, and ROC-AUC = 0.966 for strong DRD3 selectivity, whereas the receptor-independent ablation approached chance-level discrimination. These results establish protein-language-model-conditioned molecular graph learning as a scalable strategy for GPCR bioactivity prediction while identifying chemotype novelty, external-domain transfer, and richer ligand-receptor interaction representations as key remaining challenges.

bioinformatics↗

bioRxiv2026

NetPolicy-RL: Network-Informed Offline Reinforcement Learning for Pharmacogenomic Drug Prioritization

Large-scale pharmacogenomic screens provide extensive measurements of drug response across diverse cancer cell lines; however, most computational approaches emphasize point-wise sensitivity prediction or static ranking, which are poorly aligned with practical decision-making, where only a limited number of candidate drugs can be tested. We propose NetPolicy-RL, a biologically informed and decision-centric framework for pharmacogenomic drug prioritization that integrates network diffusion modeling with offline reinforcement learning. Drug selection for each cell line is formulated as an offline contextual bandit problem, enabling direct optimization of ranking quality rather than surrogate regression objectives. Mechanistic biological context is incorporated by propagating drug targets over curated interaction networks (STRING and Reactome) using random walk with restart, and combining the resulting diffusion profiles with cell-specific molecular importance derived from multi-omics data to compute network disruption scores. These biologically grounded signals are integrated with normalized drug response measurements to construct a joint state representation, which is optimized using an offline actor-critic architecture. Across held-out test splits, NetPolicy-RL consistently outperforms global ranking heuristics and learning-to-rank baselines, achieving statistically significant improvements in per-cell Normalized Discounted Cumulative Gain (NDCG@10) and substantial reductions in per-cell regret. Relative to GlobalTopK, the policy improves NDCG@10 for 88.7% of cell lines, while improvements exceed 95% compared with LambdaMART and regression-to-ranking baselines. Ablation analyses show that neither empirical response signals nor network-derived features alone are sufficient, and that their integration yields the most robust performance. Overall, this study demonstrates that combining mechanistic network biology with offline policy learning provides an effective and interpretable framework for drug prioritization in precision oncology.

bioinformatics↗

bioRxiv2025

Detection of Pre-Existing Immunity to Bacterial Cas9 Proteins in People with Cystic Fibrosis

Cystic fibrosis (CF) is caused by homozygous mutations in the cystic fibrosis transmembrane conductance regulator (CFTR) gene, resulting in multi-organ dysfunction and decreased lifespan and quality of life. A durable cure for CF will likely require a gene therapy approach to correct CFTR. Rapid advancements in genome editing technologies such as CRISPR/Cas9 have already resulted in successful FDA approval for cell-based gene editing therapies, providing new therapeutic avenues for many rare diseases. However, immune responses to gene therapy delivery vectors and editing tools remain a challenge, especially for strategies targeting complex in vivo tissues such as the lung. Previous findings in non-CF healthy individuals reported pre-existing antibody and T cell dependent immune responses to recombinant Cas9 proteins, suggesting potential additional obstacles for incorporation of CRISPR/Cas9 technologies in gene therapies. To determine if pre-existing immunity to Cas9 from S. aureus or S. pyogenes was present or augmented in people with CF (PwCF), anti-Cas9 IgG levels and Cas9-specific T cell responses were determined from peripheral blood samples of PwCF and non-CF healthy controls. Overall, non-CF controls and PwCF displayed evidence of pre-existing antibody and T cell responses to both S. aureus and S. pyogenes Cas9, although there were no significant differences between the two populations. However, we observed global changes in activation of Th1 and CD8 T cell responses as measured by IFN{gamma} and TNF that warrant further investigation and mechanistic understanding as this finding has implications not only for CRISPR/Cas9 gene therapy for PwCF, but also for protection against infectious disease.

immunology↗

bioRxiv2023

MEK1/2 inhibition decreases pro-inflammatory responses in macrophages from people with cystic fibrosis and mitigates severity of illness in experimental murine methicillin-resistant Staphylococcus aureus infection

Chronic pulmonary bacterial infections and associated inflammation remain a cause of morbidity and mortality in people with cystic fibrosis (PwCF) despite new modulator therapies. Therapies targeting host factors that dampen detrimental inflammation without suppressing immune responses critical for controlling infections remain limited, while the acquisition of antibiotic resistance bacterial infections is an increasing global problem, and a significant challenge in CF. Pharmacological compounds targeting the mammalian MAPK proteins MEK1 and MEK2, referred to as MEK1/2 inhibitor compounds, have potential combined anti-microbial and anti-inflammatory effects. Here we examined the immunomodulatory properties of MEK1/2 inhibitor compounds PD0325901, trametinib, and CI-1040 on CF innate immune cells. Human CF macrophage and neutrophil phagocytic functions were assessed by quantifying phagocytosis of serum opsonized pHrodo red E. coli, Staphylococcus aureus, and zymosan bioparticles. MEK1/2 inhibitor compounds reduced CF macrophage pro-inflammatory cytokine production without impairing CF macrophage or neutrophil phagocytic abilities. Wild-type C57BL6/J and Cftrtm1kth (F508del homozygous) mice were used to evaluate the in vivo therapeutic potential of PD0325901 compared to vehicle treatment in an intranasal methicillin-resistant Staphylococcus aureus (MRSA) infection with the community-acquired MRSA strain USA300. In both wild-type and CF mice, PD0325901 reduced infection related weight loss compared to vehicle treatment groups but did not impair clearance of bacteria in lung, liver, or spleen 1 day after infection. In summary, this study provides the first data evaluating the therapeutic potential of MEK1/2 inhibitor to modulate CF immune cells, and demonstrates that MEK1/2 inhibitors dampen pro-inflammatory responses without impairing host defense mechanisms mediating pathogen clearance.

immunology↗

bioRxiv2021

Effect of A-tract-mediated linker histone orientation on trinucleosome compaction

Linker histones (LH) have been shown to preferentially bind to AT-rich DNA, particularly A-tracts, contiguous stretches of adenines. Using spFRET (single pair Forster/Fluorescence Resonance Energy Transfer), we recently found that the globular domain (gH) of Xenopus laevis H1.0b LH orients towards A-tracts on the linker-DNA (L-DNA) while binding on-dyad in LH:mononucleosome complexes. Here, we investigate the impact of this A-tract-mediated orientation of the gH on the compaction of higher-order structures by studying trinucleosomes as minimal models for chromatin. Two 600 bp DNA sequences were constructed, each containing three consecutive Widom 601 core sequences connected by about 40 bp L-DNA but differing in the positioning of A-tracts on either the outer or the inner L-DNAs flanking the first and third Widom 601 sequences. The two inner L-DNAs were fluorescently labelled at their midpoints. Trinucleosomes were reconstituted using the doubly labelled DNA, core histone octamers and H1.0b. SpFRET was performed for a range of NaCl concentrations to measure the compaction and whether gH orientations affected the stability of the trinucleosomes to salt-induced dissociation. While the LH compacted the trinucleosomes, the extent of compaction and the stability were similar for the two DNA sequences. Modeling constrained by the measured FRET efficiency suggests that the structures adopted by the trinucleosomes correspond to the standard zig-zagged two-helical start arrangement with the first and third nucleosomes stacked on top of each other. In this arrangement, the first and third LHs are insufficiently close to interact and affect compaction. Thus, despite differences in the positioning of the A-tracts in the sequences studied, LH binding compacts the corresponding trinucleosomes similarly. Why it mattersThe compaction and three-dimensional structure of chromatin affect the exposure of the DNA and thus regulate gene expression. Linker histone proteins bind to nucleosomes and thereby contribute to chromatin compaction. We here investigated whether the DNA A-tract-mediated orientation of a linker histone globular domain affects chromatin structure by using a trinucleosome as a minimal model for chromatin. Our observations suggest that the trinucleosome structure and compaction are robust against differences in linker histone globular domain orientations. eTOC blurbWe investigate whether DNA sequences, such as adenine-tracts, and sequence-induced linker histone reorientation affect chromatin structure. Using trinucleosomes as model systems for chromatin, we demonstrate that the chromatin structure and compaction are robust to the studied DNA sequence differences and sequence-induced linker histone orientation.

biophysics↗

bioRxiv2020

DNA-sequence dependent positioning of the linker histone in a chromatosome: a single-pair FRET study

Linker histones (LH) bind to nucleosomes with their globular domain (gH) positioned in either an on- or an off-dyad binding mode. Here, we study the effect of the linker DNA (L-DNA) sequence on the binding of a full-length LH, Xenopus laevis H1.0b, to a Widom 601 nucleosome core particle (NCP) flanked by two 40 bp long L-DNA arms, by single-pair FRET spectroscopy. We varied the sequence of the 11 bp of L-DNA adjoining the NCP on either side, making the sequence either A-tract, purely GC, or mixed, with 64% AT. The labelled gH consistently exhibited higher FRET efficiency with the labelled L-DNA containing the A-tract, than that with the pure-GC stretch, even when the stretches were swapped. However, it did not exhibit higher FRET efficiency with the L-DNA containing 64% AT-rich mixed DNA when compared to the pure-GC stretch. We explain our observations with a model that shows that the gH binds on-dyad and that two arginines mediate recognition of the A-tract via its characteristically narrow minor groove. To investigate whether this on-dyad minor groove-based recognition was distinct from previously identified off-dyad major groove-based recognition, a nucleosome was designed with A-tracts on both the L-DNA arms. One A-tract was complementary to thymine and the other to deoxyuridine. The major groove of the thymine-tract was lined with methyl groups that were absent from the major groove of the deoxyuridine tract. The gH exhibited similar FRET for both these A-tracts, suggesting that it does not interact with the thymine methyl groups exposed on the major groove. Our observations thus complement previous studies that suggest that different LH isoforms may employ different ways of recognizingff AT-rich DNA and A-tracts. This adaptability may enable the LH to universally compact scaffold-associated regions and constitutive heterochromatin, which are rich in such sequences. Statement of SignificanceLinker histones (LHs) associate with the smallest repeat unit of chromatin, the nucleosome. They have been observed to have affinity for AT-rich DNA, which is found in constitutive heterochromatin and scaffold-associated regions (SAR), which could explain how the LHs can compact such parts of the chromatin. How the LH recognizes such sequences is poorly understood. Using single-pair FRET and modelling, we provide experimental evidence of DNA-sequence-induced changes in the orientation of a LH bound to a nucleosome, and thereby reveal a new mechanism by which the LH can recognize A-tract sequences that are abundantly present in the SAR. Our results show that, depending on how the LH associates with the nucleosome, it can employ more than one mechanism to recognize AT-rich DNA.

biophysics↗

bioRxiv2017

Neuronal glutamate transporters control dopaminergic signaling and compulsive behaviors

There is an ongoing debate on the contribution of the neuronal glutamate transporter EAAC1 to the onset of compulsive behaviors. Here we use behavioral, electrophysiological, molecular and viral approaches in male and female mice to identify the molecular and cellular mechanisms by which EAAC1 controls the execution of repeated motor behaviors. Our findings show that in the striatum, a brain region implicated with movement execution, EAAC1 limits group I metabotropic glutamate receptor (mGluRI) activation, facilitates D1 dopamine receptor (D1R) expression and ensures long-term synaptic plasticity. Blocking mGluRI in slices from mice lacking EAAC1 restores D1R expression and synaptic plasticity. Conversely, activation of intracellular signaling pathways coupled to mGluRI in D1R-expressing striatal neurons of mice expressing EAAC1 leads to reduced D1R expression and increased stereotyped movement execution. These findings identify new molecular mechanisms by which EAAC1 can shape glutamatergic and dopaminergic signals and control repeated movement execution.\n\nSIGNIFICANCE STATEMENTGenetic studies implicate Slc1a1, a gene encoding the neuronal glutamate transporter EAAC1, with obsessive-compulsive disorder (OCD). EAAC1 is abundantly expressed in the striatum, a brain region that is hyperactive in OCD. What remains unknown is how EAAC1 shapes synaptic function in the striatum. Our findings show that EAAC1 limits activation of metabotropic glutamate receptors (mGluRI) in the striatum and, by doing so, it promotes D1R expression. Targeted activation of signaling cascades coupled to mGluRI in mice expressing EAAC1 reduces D1R expression and triggers repeated motor behaviors in mice. These findings provide new information on the molecular basis of OCD and suggest new avenues for its treatment.

neuroscience↗