Biology subjects

Su, W.-C.

Publications and source records attributed to Su, W.-C..

5 recordsLinked to original sources

bioRxiv2026

Assessing the Reliability of LLM-Generated Phenotype-Genotype Associations Through External Validation

BackgroundPhenotype-genotype associations underpin precision medicine by enabling disease prevention, early diagnosis, risk stratification, therapeutic target discovery, and personalized treatment. However, the rapid growth of scientific evidence has made manual curation of these associations increasingly labor-intensive, time-consuming, and incomplete. Large Language Models (LLMs) offer a potential path to scalable genomic generation and synthesis of this knowledge, but their ability to accurately identify phenotype-genotype associations and the extent to which these outputs are supported by established genomic knowledge bases remain unclear. Materials and MethodsFour LLMs, Claude Sonnet 4.6, DeepSeek V4 Flash, Gemini 3 Flash Preview, and GPT-5.5, were benchmarked on six zero-shot task categories covering forward and reverse phenotype-gene and phenotype-SNP generation. A total of 4,196 associations were identified from curated inputs and evaluated through a multistage external verification pipeline comprising phenotype normalization, ontology mapping, genomic identifier validation against Ensembl, and evidence verification using both the GWAS Catalog and OMIM. Associations were assigned a fused evidence level of strong, moderate, weak, or none. ResultsOverall, 74.19% of generated associations were matched to at least one external genomic knowledge base; 9.15% received strong support and 54.46% moderate support. Phenotype-gene associations were more verifiable than phenotype-SNP associations (strong or moderate: 67.19% vs 54.06%). Among existing associations, Claude Sonnet 4.6 achieved the highest overall strong or moderate rate (69.2%), followed by GPT-5.5 (65.1%), DeepSeek V4 Flash (61.7%), and Gemini 3 Flash Preview (56.9%). ConclusionLLMs can support scalable generation of candidate phenotype-genotype associations. Performance varied substantially by relation type and was lower for SNP-level and rare disease associations, highlighting both the limitations of current genomic resources and the need for rigorous validation pipelines.

bioinformatics↗

bioRxiv2026

Divergent landscapes of positive and negative selection signatures across residue-resolved human-virus protein-protein interaction interfaces

Virus-targeted host proteins evolve under dual selective pressures. Negative selection preserves within-host interactions, while positive selection promotes adaptive changes to evade viral engagement. Viral and endogenous within-host partners can compete for binding, bringing distinct pressures together on the same interaction interface. Yet, the spatial organization of distinct selective pressures across virus-targeted host proteins, and how such pressures manifest across diverse interaction contexts, remains largely unknown. Here, we integrate an evolutionarily annotated map of human-virus protein-protein interactions (PPIs) with intra-protein residue-residue contact maps to probe the spatial organization of residue-level selective pressures across PPI interfaces of virus-targeted host proteins. Across all PPI interfaces collectively, we find that residues under positive selection are spatially clustered, whereas those under negative selection are broadly dispersed, with additional spatial segregation between positive and strongly negatively selected sites. Moreover, while positive selection is unevenly distributed across interfaces bound exclusively by viral proteins (exogenous-specific), they are more uniformly distributed across interfaces shared between viral and within-host partners (mimic-targeted), suggesting that adaptive pressure from viral targeting acts on the entire mimic-targeted interface, whereas it acts on only a subset of the exogenous-specific interface. Strikingly, clustering of positively selected residues is more pronounced between mimic-targeted and other interface types than within exogenous-or endogenous-specific interfaces alone, suggesting that mimic-targeted interfaces may serve as focal points of adaptive evolution. Overall, our multiscale framework of PPI interfaces and residue-level contacts reveals heterogeneous, context-dependent landscapes of selective pressures across virus-targeted host proteins, providing a high-resolution view of how adaptation and constraint are intricately balanced and coordinated within the host.

bioinformatics↗

bioRxiv2021

Plasma Contains Ultra-short Single-stranded DNA in Addition to Nucleosomal cfDNA

Plasma cell-free DNA is a widely used biomarker for diagnostic screening. We introduce uscfDNA-seq, a single-stranded cell-free DNA NGS pipeline, which bypasses previous limitations to reveal a novel population of ultrashort single-stranded cell-free DNA in plasma with a modal size of 50nt. This species of cfDNA aligns predominantly to the nuclear genome and could potentially be used for novel biomarker discovery.

genomics↗

bioRxiv2020

Recurrent Dynamics of Rupture Transitions of Single Giant Vesicles at Solid Surfaces

Single giant vesicles (GVs) rupture spontaneously from their salt-laden suspension onto solid surfaces. At hydrophilic surfaces, they rupture via a recurrent burst-heal dynamics: during burst, single pores nucleate at the contact boundary of the adhering vesicles facilitating asymmetric spreading and producing a "heart" shaped membrane patch. During the healing phase, the competing pore closure produces a daughter vesicle. At hydrophobic surfaces, by contrast, the GVs rupture via a distinctly different, yet recurrent, bouncing ball rhythm: Rendered tense by the substrate interactions, GVs porate and spread monomolecular layer on the hydrophobic surface in a symmetric manner. Here too, the competition from pore closure produces a daughter vesicle, which re-engages with the substrate. In both cases, the pattern of burst-reseal events repeats multiple times splashing and spreading the vesicular fragments as bilayer patches at the solid surface in a pulsatory manner. These remarkable recurrent dynamics arise not because of the elastic properties of the solid surface but because the competition between membrane spreading and pore healing, prompted by the surface-energy dependent adhesion, determine the course of the topological transition. STATEMENT OF SIGNIFICANCEGiant lipid vesicles adhering to a solid surface experience strong mechanical stresses. The contacting membrane segment loses thermal fluctuations and accumulates mechanical tension, the equilibration of which can give rise to global shape changes, lipid phase separation, and traction forces. Beyond a threshold tension, vesicles porate, unravel, and spread. Here, we find that a competition from pore-healing can make rupture iterative, rather than a single all-or-nothing event. During burst, single pores expand, spreading a lipid bilayer on the hydrophilic surface and a monolayer on the hydrophobic one. During heal, pore-healing can produce daughter vesicles. This burst-reseal event reiterates "splashing" portions of single vesicles at the solid surface and "bouncing" the remainder as a secondary vesicle in multiple steps.

biophysics↗

bioRxiv2020

Improved genetically encoded near-infrared fluorescent calcium ion indicators for in vivo imaging

Near-infrared (NIR) genetically-encoded calcium ion (Ca2+) indicators (GECIs) can provide advantages over visible wavelength fluorescent GECIs in terms of reduced phototoxicity, minimal spectral cross-talk with visible-light excitable optogenetic tools and fluorescent probes, and decreased scattering and absorption in mammalian tissues. Our previously reported NIR GECI, NIR-GECO1, has these advantages but also has several disadvantages including lower brightness and limited fluorescence response compared to state-of-the-art visible wavelength GECIs, when used for imaging of neuronal activity. Here, we report two improved NIR GECI variants, designated NIR-GECO2 and NIR-GECO2G, derived from NIR-GECO1. We characterized the performance of the new NIR GECIs in cultured cells, acute mouse brain slices, and C. elegans and Xenopus laevis in vivo. Our results demonstrate that NIR-GECO2 and NIR-GECO2G provide substantial improvements over NIR-GECO1 for imaging of neuronal Ca2+ dynamics.

molecular biology↗