Sergei Maslov

This paper introduces Competitive Dimerization Networks (CDNs) for chemical learning, demonstrating their ability to perform multiclass classification through in vitro directed evolution without digital hardware.

This study uses CRISPR base editing to create HTT protein isoforms resistant to caspase-6 cleavage, reducing toxicity and neurodegeneration in a rodent HD model.

The paper shows that survivorship bias, not biological changes, explains age-dependent extinction patterns in fossil genera using a one-parameter model.

The paper shows that single-cell heterogeneity in IFN induction potential is heritable and driven by baseline variation in JNK and AP-1 signaling states.

The paper compares fitness of microbial metabolic strategies in fluctuating environments, finding sequential utilizers gain an advantage in diverse communities.

This study introduces a consumer-resource model to analyze higher-order interactions in auxotroph communities and shows how these interactions enhance resilience to resource fluctuations.

The paper shows that resource ratio fluctuations drive the evolution of microbial metabolic strategies, with balanced environments favoring specialists and imbalanced ones favoring generalists.

The paper introduces Living Neurosheets, modular neural systems that enable scalable, bio-inspired neural architectures with spatial control and multimodal integration.

This study shows that protein language models capture both structural and functional dependencies from sequence data alone through epistasis analysis.

The paper develops a model to predict serial dilution dynamics in a synthetic human gut microbiome using metabolomics data and validates it with high correlation.

The paper identifies intrinsic OASL expression as a determinant of IFN induction potential in influenza A virus-infected cells using scRNA-seq analysis.

SPLICER improves exon skipping by editing both splice acceptor and donor sites, reducing aberrant splicing and enabling therapeutic applications in Alzheimer's disease.

noSpliceVelo infers gene expression dynamics without requiring unspliced and spliced transcript separation, performing comparably or better than scVelo on multiple datasets.

DR-BERT is a protein language model pretrained on unannotated proteins to predict disordered regions, outperforming existing methods without using explicit evolutionary or biophysical features.

FUN-PROSE predicts condition-specific gene expression in fungi using promoter sequences and transcription factor data, achieving high correlation with observed expression.

A RoBERTa-style Transformer, pre-trained on 450K Swiss-Prot sequences in 18 hours on 4 V100 GPUs, rivals DeepFam at family classification and beats PIPR at human protein-protein interaction prediction.

Open-source KBase platform unites reference genomes and biochemistry, over 70 apps, Jupyter-based Narratives with provenance, and a Docker SDK for reproducible plant and microbial systems biology.

The paper studies a theoretical model of heteropolymers to show how natural selection-like processes can emerge, leading to a reduction in information entropy and the survival of a subset of polymer sequences.

The paper analyzes 134 coexpression networks from Arabidopsis microarray datasets to show context-dependent gene coexpression patterns and introduces pan- and core-network concepts.

The paper models diversity waves in ecosystems caused by population collapses, showing cyclic patterns and a scale-free distribution with a universal exponent.