Biology subjects

Dias, R.

Publications and source records attributed to Dias, R..

12 recordsLinked to original sources

bioRxiv2026

LifeSciBench: Evaluating Language Models on Realistic, Expert-Level Tasks in the Life Sciences

We introduce LifeSciBench, a benchmark of 750 expert-authored tasks designed to evaluate whether language models can handle realistic life science research work. The majority of existing life sciences benchmarks have a narrow scope or are purely knowledge-based, and therefore fail to capture the complexity of real-world research, which often involves ambiguities and requires the accurate execution of multiple dependent judgment calls. Additionally, almost all existing benchmarks span at best a small collection of subdomains within the life sciences; there is at present no existing life sciences benchmark with both the requisite breadth and depth required to convincingly measure proficiency in real-world professional research settings. LifeSciBench addresses this gap by spanning seven representative scientific workflows and seven life science domains, with each constituent task paired with a human expert-written rubric. Across five frontier and domain-specialized models, GPT-Rosalind performs best, with a task-weighted mean normalized rubric score of 0.576 and a task-weighted response pass rate of 36.1% (response-level values are first averaged within each task, and the resulting task-level values are then averaged with equal weight). LifeSciBench remains unsaturated, with 171 tasks (22.8%) having no observed passing response from any evaluated model and 261 tasks (34.8%) having a best-model pass rate below 20%. LifeSciBench therefore serves as a high-resolution evaluation of practical scientific reasoning and operational decision-making in the life sciences.

bioinformatics↗

bioRxiv2026

SpaceBio Knowledge Hub: A LiteratOmics Platform for Microgravity and Space Biology Research

Space biology literature is growing exponentially. Existing infrastructure has not kept pace with organizing, synthesizing, and disseminating this knowledge. We present SpaceBio SpaceBio Knowledge Hub (www.spacebio.space), an integrated digital ecosystem that combines artificial intelligence, real-time data integration, and open-access infrastructure to advance research, education, and collaboration in microgravity, space biology and space exploration. The platform applies AI-driven approaches including natural language processing, machine learning, and automated content generation to construct a semantic atlas of the field. The atlas reveals the hierarchical thematic organization underlying microgravity-induced biological responses, space mission infrastructure, planetary science, and astrobiology. As part of this effort, SpaceBio is moving toward the construction of a LiteratOmics framework for microgravity, and space biology a systematic, AI-enabled approach to mining, integrating, and structuring the primary literature generated by omics-driven spaceflight research, treating the scientific literature itself as a navigable data layer alongside genomic, transcriptomic, and proteomic datasets. Built on a scalable, cloud-based architecture with a user-centered interface, SpaceBio supports literature exploration, data integration, and knowledge discovery for researchers, educators, students, industry partners, and citizen scientists. The platform also functions as a community-building ecosystem. It integrates hands-on research initiatives, AI-generated educational content, pilot data science projects, and social responsibility programs that broaden participation without compromising scientific rigor. AI-enabled digital environments can transform fragmented literature into a navigable knowledge landscape. SpaceBio accelerates research productivity, strengthens STEM education, and supports the global space life sciences community as human space exploration enters in the most ambitious era.

scientific communication and education↗

bioRxiv2026

Explainable AI identifies recombination and chromatin environment as key predictors of subgenome evolution in maize and Brassica

Polyploidy, or whole genome duplication, reshapes genomes through biased gene loss and regulatory rewiring, yet the drivers of biased fractionation among subgenomes remain unclear. Using maize and Brassica rapa as model allopolyploids, we compiled 60 genomic and epigenomic features in maize and 45 in B. rapa and constructed supervised machine-learning models to classify genes by subgenome identity. In both species, eXplainable Artificial Intelligence (XAI) approaches identified recombination rate as the most influential and highly interconnected feature despite nonsignificant mean differences between subgenome groups. Chromosome location, transposon density, and chromatin-associated features, including the proximity of accessible chromatin regions to genes and active histone marks such as H3K9ac and H3K27ac, consistently ranked among the top contributors to subgenome classification. XAI-derived co-variation and interaction networks further revealed recombination rate as the central node connecting significant features. Together, these results highlight recombination and chromatin environment as major predictors of subgenome divergence in polyploid genomes. TeaserRecombination and chromatin environment are key predictors of unequal gene loss and subgenome dominance after polyploidy.

plant biology↗

bioRxiv2026

QuadCleave: Enzymatic cleavage and analysis of DNA G-quadruplexes.

Prediction of G-quadruplexes (G4), noncanonical DNA secondary structures containing guanine tetrads, can have profound relevance across cancer, neurodegenerative and other genetic diseases, but require experimental validation by instrument intensive methods including CD spectroscopy, fluorescence, NMR, or crystallography. We report a novel, inexpensive, efficient and scalable 2-enzyme system for confirmation of G4 formation and copy number repeat count. First, QuadCleave, in which Mismatch Endonuclease I was used to repeatably, site-specifically and selectively cleave oligodeoxynucleotides containing exclusively G4-forming sequences. Second, HaeIII enzyme identified G4-forming sequences that can also generate GG/CC base pairs, and a sequences cleavage by both established it in C9ORF72, the most mutated gene in amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD). These findings establish ubiquitously accessible experimental validation of G4 forming sequences and their copy number repeat assessment, with specificity to C9ORF72 oligodeoxynucleotides whose repeat expansion has critical clinical implications in a subpopulation of ALS and FTD patients.

molecular biology↗

bioRxiv2026

Integrating targeted genome mining and structure-guided modeling reveals unexplored 7-deazapurine-containing pathways

7-deazapurines are nucleoside analogs that play key roles in nucleic acid modification and can serve as building blocks for diverse, bioactive secondary metabolites. Despite their biological significance, their biosynthetic diversity, distribution, and enzymatic determinants of structural diversification remain poorly understood. Here, we leverage large-scale targeted genome mining, phylogenetic, and network analysis to explore 7-deazapurine-containing pathways across [~]2 million bacterial genomes. We identified over 900 candidate biosynthetic gene clusters (BGCs), grouped into more than 100 families, most of which remain uncharacterized. These GATOR-GC-predicted BGCs were predominantly found in Streptomyces. We then examined enzyme-substrate interactions in three representative pathways: (i) peptidyl-deazapurines, (ii) huimycin, and (iii) dapiramicin A. Molecular docking and molecular dynamics (MD) simulations recapitulated known enzyme-substrate interactions and highlighted candidate catalytic residues governing amide bond formation, methylation, and glycosylation. Using this genome- and structure-guided framework, we identified a candidate BGC for dapiramicin A and proposed tailoring steps, including scaffold methylation and deoxy-sugar formation. These findings expand the known diversity of 7-deazapurine-containing BGCs and demonstrate how integrating genome mining with structural modeling can link BGCs to chemical function, providing a foundation for discovering and characterizing 7-deazapurine-containing secondary metabolites. Graphical abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=79 SRC="FIGDIR/small/718813v1_ufig1.gif" ALT="Figure 1"> View larger version (29K): org.highwire.dtl.DTLVardef@c00feforg.highwire.dtl.DTLVardef@156468forg.highwire.dtl.DTLVardef@1326e90org.highwire.dtl.DTLVardef@1f8d57b_HPS_FORMAT_FIGEXP M_FIG C_FIG

bioinformatics↗

bioRxiv2025

Thermally activated irreversible homogenization of G-quadruplexes in an ALS/FTD-associated nucleotide expansion

A significant proportion of familial amyotrophic lateral sclerosis (ALS) and frontotemporal dementia (FTD) cases exhibit a substantial copy number expansion of the hexanucleotide GGGGCC/GGCCCC sequence in the C9ORF72 gene. The GGGGCC sequence forms a non-canonical DNA structure called a G-quadruplex (G4) which has been associated with the disease states and with nucleic acid condensate formation. G4s can fold into various topologies, which can differentially impact fidelity of DNA synthesis. However, how G4 conformational heterogeneity and its regulation impact hexanucleotide repeat expansion is unclear, and important clues may lie in the thermodynamic properties of different G4 topologies. Here, we use temperature-swept CD spectroscopy to observe configurational homogenization of an initially heterogeneous population of G4s over a small range of temperatures, demonstrating thermally activated behavior. We further show that this reaction is irreversible, since subsequent temperature sweeps do not show CD shifts from non-parallel to parallel G4 topologies. Finally, we provide an analytical theory based on a two-state thermodynamic model which is compatible with experimental evidence, and we discuss alternate mechanisms for the homogenization transition. These findings suggest that kinetic regulation of non-canonical DNA structures may play a role in cellular homeostasis or disease pathogenesis. SIGNIFICANCEThe GGGGCC repeats in the C9ORF72 gene expand in copy number in certain neurodegenerative diseases, forming non-canonical DNA structures called G-quadruplexes (G4) which are associated with the pathological state. However, why the repeat expansion occurs is not known, and a key may lie in the thermodynamic stability of certain G4 conformations. Here, we use CD spectroscopy to experimentally report thermally activated heat-induced G4 conformation homogenization, from a heterogeneous population to the parallel configuration. We derive an analytical biophysical theory which is compatible with this experimental observation, which is shown to be irreversible. Our in vitro tuning of the free energy landscape that modulates G4 conformational fidelity motivates a search for possible in vivo enzymatic regulators.

biophysics↗

bioRxiv2025

InteracTor: Feature Engineering and Explainable AI for Profiling Protein Structure-Interaction-Function Relationships

Characterizing protein families structural and functional diversity is essential for understanding their biological roles. Traditional analyses often focus on primary and secondary structures, which may not fully capture complex protein interactions. Here we introduce InteracTor, a novel toolkit that extracts multimodal features from protein three-dimensional (3D) structures, including interatomic interactions like hydrogen bonds, van der Waals forces, and hydrophobic contacts. By integrating Explainable AI (XAI) techniques, we quantified the importance of the extracted features in the classification of protein structural and functional families. InteracTors interpretable features enable mechanistic insights into the determinants of protein structure, function, and dynamics, offering a transparent means to assess their predictive power within machine learning models. Interatomic interaction features extracted by InteracTor demonstrated superior predictive power for protein family classification compared to features based solely on primary or secondary structure, revealing the importance of considering specific tertiary contacts in computational protein analysis. This work provides a robust framework for future studies aiming to enhance the capabilities of models for protein function prediction and drug discovery. AUTHOR SUMMARYInteracTor is a computational toolkit designed to enhance our understanding of protein structure and function by focusing on three-dimensional (3D) structural interactions. Unlike traditional approaches that primarily rely on sequence or secondary structure data, InteracTor extracts biologically meaningful features such as hydrogen bonds, van der Waals forces, and hydrophobic contacts, which are critical for protein stability and dynamics. By integrating these features into machine learning models alongside explainable AI methods, InteracTor provides interpretable insights into how specific structural interactions influence protein behavior. Our results demonstrate that tertiary structure features significantly improve the accuracy of protein family classification compared to sequence-based methods alone, underscoring the importance of considering 3D interactions in computational protein analyses. The toolkits modular design makes it adaptable for diverse applications, including drug discovery and protein engineering. In a broader context, InteracTor bridges the gap between computational biology and practical applications in medicine and biotechnology by offering a transparent and robust framework for analyzing proteins at a molecular level. This work represents a step forward in leveraging structural data to advance predictive modeling and biological discovery.

bioinformatics↗

bioRxiv2024

InteracTor: A new integrative feature extraction toolkit for improved characterization of protein structural properties

Understanding the structural and functional diversity of protein families is crucial for elucidating their biological roles. Traditional analyses often focus on primary and secondary structures, which include amino acid sequences and local folding patterns like alpha helices and beta sheets. However, primary and secondary structures alone may not fully represent the complex interactions within proteins. To address this limitation, we developed a new algorithm (InteracTor) to analyze proteins by extracting features from their three-dimensional (3D) structures. The toolkit extracts interatomic interaction features such as hydrogen bonds, van der Waals interactions, and hydrophobic contacts, which are crucial for understanding protein dynamics, structure, and function. Incorporating 3D structural data and interatomic interaction features provides a more comprehensive understanding of protein structure and function, potentially enhancing downstream predictive modeling capabilities. By using the extracted features in Mutual Information scoring (MI), Principal Component Analysis (PCA), t-distributed Stochastic Neighbor Embedding (t-SNE), Uniform Manifold Approximation and Projection (UMAP), and hierarchical clustering analysis as use cases, we identified clear separations among protein structural families, highlighting distinct functional aspects. Our analysis revealed that interatomic interaction features were more informative than protein secondary structure features, providing insights into potential structural and functional properties. These findings underscore the significance of considering tertiary structure in protein analysis, offering a robust framework for future studies aiming at enhancing the capabilities of models for protein function prediction and drug discovery.

bioinformatics↗

bioRxiv2024

Limitations of Current Machine-Learning Models in Predicting Enzymatic Functions for Uncharacterized Proteins

Thirty to seventy percent of proteins in any given genome have no assigned function and have been labeled as the protein "unknome". This large knowledge shortfall is one of the final frontiers of biology. Machine-Learning (ML) approaches are enticing, with early successes demonstrating the ability to propagate functional knowledge from experimentally characterized proteins. An open question is the ability of machine-learning approaches to predict enzymatic functions unseen in the training sets. Using a set of Escherichia coli unknowns, we evaluated the current state-of-the-art machine-learning approaches and found that these methods currently lack the ability to integrate scientific reasoning into their prediction algorithms. While human annotators can leverage the plethora of genomic data in making plausible predictions into the unknown, current ML methods not only fail to make novel predictions but also make basic logic errors in their predictions. This underscores the need to include assessments of prediction uncertainty in model output and to test for hallucinations (logic failures) as a part of model evaluation. Explainable AI (XAI) analysis can be used to identify indicators of prediction errors, potentially identifying the most relevant data to include in the next generation of computational models. Article SummaryMany proteins in any genome, ranging from 30% to 70% of the genome, lack an assigned function. This knowledge gap limits the full use of the vast available genomic data. Machine learning has shown promise in transferring functional knowledge within isofunctional families, but it largely fails to predict novel functions not seen in its training data. Understanding these failures can guide the development of better machine-learning methods to help experts make accurate functional predictions for uncharacterized proteins.

bioinformatics↗

bioRxiv2024

Metabolic and age-associated epigenetic barriers during direct reprogramming of mouse fibroblasts into induced cardiomyocytes

Heart disease is the leading cause of mortality in developed countries novel regenerative procedures are warranted for improving patients well fare. Direct cardiac conversion (DCC) can create induced cardiomyocytes (iCMs) and besides holding great promise, lacks clinical effectiveness for cardiac regeneration as metabolic and age-associated barriers remains elusive. Here, we identify by histone post-translational analysis that DCC triggers major age-dependent alterations in the epigenetic landscape. Metabolomics revealed decrease abundance of anabolic metabolites related to TCA cycle and glutaminolysis and profound mitochondrial network remodeling, with increased elongation and interconnectivity and reliance in mitochondrial respiration in iCMs. Importantly, adult-derived iCMs present increase accumulation of oxidative stress in the mitochondria and pharmacological activation of mitophagy increase DCC of adult fibroblasts in vitro. Metabolic modulation in vitro and dietary manipulations in vivo improves DCC efficiency and are accompanied by significant alterations in the histone acetylation and methylation landscape and mitochondria homeostasis. Our study provides evidence that metaboloepigenetics as a direct role in cell fate transitions driving direct cardiac conversion into iCMs, highlighting the potential use of metabolic modulation in increasing cardiac regenerative strategies.

cell biology↗

bioRxiv2023

Characterizing constitutive promoters across the Proteobacteria

Although research on promoters has spanned decades, the precise prediction of promoter activity from DNA sequence remains a challenge even in model organisms. Recent literature has identified important differences in the core sequence of {sigma}70 promoters across classes of Proteobacteria as well as a lack of transferability when promoters are moved from host to host. Currently, there is a need for synthetic constitutive promoters spanning a range of expression levels in species outside of Escherichia coli. Additionally, characterization data defining behavior of the same promoter across multiple species would be extremely valuable to the field. Here, we analyzed promoter activity in three classes of Proteobacteria, which enabled us to better understand the sequence elements correlated with a strong promoter in different hosts. In doing so, we identified and characterized constitutive promoters spanning a range of expression in these species for community use and described the portability of a subset of these promoters as they were moved between hosts. These promoter libraries have broad applications as predictable genetic tools to control gene expression in diverse species (1-3). This work adds to the toolkit for gene expression in non-model bacteria and is a step towards the larger goal of accurate promoter prediction in a given host from a de novo sequence.

synthetic biology↗

bioRxiv2021

Rapid, Reference-Free Human Genotype Imputation with Denoising Autoencoders

Genotype imputation is a foundational tool for population genetics. Standard statistical imputation approaches rely on the co-location of large whole-genome sequencing-based reference panels, powerful computing environments, and potentially sensitive genetic study data. This results in computational resource and privacy-risk barriers to access to cutting-edge imputation techniques. Moreover, the accuracy of current statistical approaches is known to degrade in regions of low and complex linkage disequilibrium. Artificial neural network-based imputation approaches may overcome these limitations by encoding complex genotype relationships in easily portable inference models. Here we demonstrate an autoencoder-based approach for genotype imputation, using a large, commonly-used reference panel, and spanning the entirety of human chromosome 22. Our autoencoder-based genotype imputation strategy achieved superior imputation accuracy across the allele-frequency spectrum and across genomes of diverse ancestry, while delivering at least 4-fold faster inference run time relative to standard imputation tools.

bioinformatics↗