Research Radar — 2026-08-07
Methods & AI
Computational
Contrastive Regulatory Embeddings Attention Model for Differential Expression Prediction with Personalized Genomes: Insights and Challenges
bioRxiv (bioinformatics) Published 2026-08-04 Preprint DOI: 10.64898/2026.08.02.742309v1
deep learning personalized genomes gene regulation expression prediction
Summary: The authors propose CREAM, a contrastive regulatory embedding attention model that decomposes expression into genetic and non-genetic components and improves prediction on training genes in simulated and GTEx data.
Why it matters: The study makes the generalization failure on unseen genes explicit while testing architectural and uncertainty-aware strategies for personalized gene regulation.
Why for Yiru: This is directly relevant to your interests in AI for genomics because it connects regulatory sequence modeling, fine-mapped variants and the limits of cross-gene generalization.
MEGA-ODE: Learning Biologically Structured and Navigable Continuous Perturbation Dynamics from Sparse Omics
bioRxiv (bioinformatics) Published 2026-08-04 Preprint DOI: 10.64898/2026.08.05.742921v1
perturbation omics graph neural ODEs drug response virtual perturbations
Summary: MEGA-ODE combines molecular-network priors, graph neural ordinary differential equations and context-adaptive mixture-of-experts routing to reconstruct sparse continuous-time omics dynamics and prioritize virtual perturbations.
Why it matters: It treats missing molecular states and interventions as a navigable dynamical-system problem, with evaluations spanning transcriptomics, proteomics, infection and differentiation.
Why for Yiru: This fits your computational biology interests by turning sparse perturbation data into interpretable trajectory forecasts and experimentally testable intervention hypotheses.
Evaluating the impact of a sample-matched reference genome on single-cell transcriptomic inferences in Plasmodium falciparum
bioRxiv (bioinformatics) Published 2026-08-04 Preprint DOI: 10.64898/2026.07.30.740912v1
single-cell RNA-seq reference genomes Plasmodium falciparum variant mapping
Summary: Using a naturally infected Mali isolate, the authors compare single-cell RNA-seq inference against the standard Pf3D7 reference and a sample-matched ML52 assembly, finding broadly concordant results for conserved genes but substantial differences for highly polymorphic var genes.
Why it matters: It shows that reference choice can be a biological confounder rather than a routine preprocessing detail when transcriptomic targets are highly variable.
Why for Yiru: This is useful for your single-cell and pathogen-genomics work because it links mapping design to the reliability of cell-state and antigenic-locus inference.
Homology-Based Variant-Effect Predictors Break Down on Cytochrome P450 Pharmacogenes
bioRxiv (bioinformatics) Published 2026-08-04 Preprint DOI: 10.64898/2026.07.30.741616v1
variant-effect prediction pharmacogenomics protein language models deep mutational scanning
Summary: The study reports that homology-based predictors perform poorly on cytochrome P450 pharmacogenes and that an embedding-based k-nearest-neighbors ensemble improves correlation with CYP2C9 deep-mutational-scanning activity, although clinical agreement remains limited.
Why it matters: It identifies substrate dependence as a possible missing feature in variant-effect prediction for multi-substrate enzymes rather than treating evolutionary conservation as a universal proxy for function.
Why for Yiru: This is a valuable model-evaluation case for your AI-biology interests because it separates benchmark improvement from clinical validity.
Biomedical discoveries
Biomedicine
Developmental reversion underlies resistance to immune checkpoint blockade in kidney cancer
bioRxiv (cancer biology) Published 2026-08-05 Preprint DOI: 10.64898/2026.08.05.743137v1
kidney cancer immune checkpoint blockade single-cell RNA-seq spatial transcriptomics
Summary: An atlas of 110 clear cell renal cell carcinoma patients combines single-cell RNA sequencing and imaging-based spatial transcriptomics to identify enrichment of early nephrogenesis programs in immune-checkpoint-resistant persister cells, with validation in an immunocompetent mouse model.
Why it matters: The findings frame acquired immunotherapy resistance as a developmental reversion state with associated Notch, injury-repair and inhibitory checkpoint programs.
Why for Yiru: This strongly matches your tumor-microenvironment interests by linking cancer-cell state, spatial niche and immune-treatment resistance.
SWI/SNF Alterations Define a Chromatin-Dependent Subtype of Urothelial Carcinoma
bioRxiv (cancer biology) Published 2026-08-05 Preprint DOI: 10.64898/2026.08.05.743022v1
urothelial carcinoma SWI/SNF chromatin remodeling HDAC inhibitors
Summary: Integrative genomic and transcriptomic analyses of 792 urothelial carcinomas, validated in TCGA-BLCA, define a BAF-altered subtype with proliferative, lineage-loss and metabolic programs and report increased sensitivity to HDAC inhibition in models and early clinical observations.
Why it matters: It connects a recurrent chromatin alteration to a biomarker-defined therapeutic vulnerability using genomic, epigenomic, functional and preliminary clinical evidence.
Why for Yiru: This is relevant to your regulatory and cancer-biology interests because it ties chromatin state to treatment response and patient stratification.
Invasion status stratifies the composition of the human pancreatic cancer perineural niche
bioRxiv (cancer biology) Published 2026-08-05 Preprint DOI: 10.64898/2026.08.05.743048v1
pancreatic cancer perineural invasion spatial transcriptomics tumor microenvironment
Summary: A spatial and single-cell atlas of the human pancreatic ductal adenocarcinoma perineural niche uses AI-based imaging transcriptomics and pathology-guided single-nucleus RNA-seq to contrast invaded and non-invaded nerve neighborhoods.
Why it matters: The study associates invaded neighborhoods with classical cancer cells, myofibroblastic CAFs and lipid-associated macrophages, while non-invaded neighborhoods contain inflammatory CAFs and B and T cells.
Why for Yiru: This directly fits your spatial biology and tumor-microenvironment interests by making invasion status a key axis of niche composition.
FLIP is essential for oncogenic KRAS-driven lung cancer
bioRxiv (cancer biology) Published 2026-08-04 Preprint DOI: 10.64898/2026.08.04.738686v1
KRAS lung cancer FLIP/CFLAR apoptosis
Summary: The authors identify FLIP/CFLAR as a dependency of KRAS-mutant lung cancer, showing in cell lines and genetically engineered mouse models that FLIP loss triggers caspase-8-dependent apoptosis and sensitizes tumors to inflammatory cytokines.
Why it matters: The proposed KRAS–ERK–FLIP axis offers a mechanistic explanation for survival under oncogenic stress and a possible biomarker-guided combination strategy.
Why for Yiru: This suits your cancer and immune-microenvironment interests by connecting oncogenic signaling with cytokine-driven cell death.
Endocytosis of ALK promotes glucose uptake in ALK-amplified neuroblastoma
bioRxiv (cancer biology) Published 2026-08-04 Preprint DOI: 10.64898/2026.08.04.742396v1
neuroblastoma ALK endocytosis cancer metabolism
Summary: In ALK-amplified neuroblastoma cells, inhibiting ALK or receptor endocytosis reduces glucose uptake by roughly 40–50%; the process requires dynamin, dynein and GLUT1 and supports growth independently of ERK MAPK and PI3K-AKT signaling.
Why it matters: It identifies receptor endocytosis as a noncanonical metabolic vulnerability distinct from the canonical downstream kinase pathways.
Why for Yiru: This is useful for your signaling and tumor-metabolism interests because it links receptor trafficking to mitochondrial glucose delivery and selective cancer growth.
The coumarin derivative X6632 is a pan-ID protein inhibitor that suppresses tumor growth by targeting cancer cells and the tumor-associated microvasculature
bioRxiv (cancer biology) Published 2026-08-04 Preprint DOI: 10.64898/2026.08.05.742008v1
ID proteins tumor microvasculature angiogenesis immune checkpoint blockade
Summary: The coumarin derivative X6632 suppresses ID-protein expression, tumor-cell functions and endothelial-cell functions in vitro, reduces pathological vascularization and tumor growth in models, and improves tumor control with immune checkpoint blockade.
Why it matters: It presents a preclinical dual-compartment strategy aimed at malignant cells and tumor-supporting vasculature rather than either compartment alone.
Why for Yiru: This complements your tumor-microenvironment interests by connecting cancer-cell stemness, angiogenesis and combination immunotherapy.
Cross-disciplinary watchlist
Other Fields
Integrated Pangenomic and Systems Biology Analyses Reveal the Genomic Basis of Virulence and Adaptation in Bipolaris sorokiniana
bioRxiv (bioinformatics) Published 2026-08-04 Preprint DOI: 10.64898/2026.07.31.741966v1
fungal pathogens pangenomics virulence systems biology
Summary: Across 19 globally distributed Bipolaris sorokiniana genomes, the authors report an open pangenome with a dynamic accessory fraction, dozens of biosynthetic gene clusters per genome and systems-level signals associated with adaptation and virulence.
Why it matters: It provides a population-scale genomic framework for studying cereal-pathogen diversification, resistance breakdown and fungicide resistance.
Why for Yiru: This broadens your comparative genomics perspective by combining pangenome structure, gene-family evolution and interactome analysis in a crop pathogen.
CARS: A General Force Field for Carotenoids
bioRxiv (bioinformatics) Published 2026-08-04 Preprint DOI: 10.64898/2026.07.31.742033v1
molecular dynamics carotenoids force fields computational chemistry
Summary: CARS is a transferable carotenoid force field compatible with AMBER that is optimized against 22,957 r2SCAN-3c reference energies and validated across glycosylated, acylated and lipid-linked carotenoids.
Why it matters: The parameterization aims to remove molecule-specific reparameterization while improving conformational energetics relative to established general and carotenoid-specific force fields.
Why for Yiru: This is a practical computational-methods resource for your molecular modeling interests because it supports simulations of chemically diverse biological pigments.
A Comprehensive Database of Simulations and Meshes of Coronary Arteries from the Fame 2 Trial
bioRxiv (bioinformatics) Published 2026-08-04 Preprint DOI: 10.64898/2026.07.30.741868v1
coronary arteries computational hemodynamics finite elements machine learning datasets
Summary: The authors release 779 reconstructed coronary-vessel meshes and corresponding unsteady Navier–Stokes simulations from the FAME 2 trial, using hexahedral meshes with identical connectivity and coronary outlet boundary conditions.
Why it matters: It addresses the scarcity of public numerical hemodynamics data and creates a standardized resource for data-driven cardiovascular modeling.
Why for Yiru: This is useful for your computational biology interests as an example of turning clinical imaging into a reusable simulation and machine-learning benchmark.
VRK1 kinase maintains an undifferentiated proliferative state in neuroblastoma tumor cells
bioRxiv (cancer biology) Published 2026-08-04 Preprint DOI: 10.64898/2026.08.05.742965v1
neuroblastoma VRK1 cell differentiation cancer stemness
Summary: Patient datasets, tissue assays, single-cell transcriptomics and model systems associate VRK1 with undifferentiated neuroblastoma; VRK1 depletion promotes differentiation-associated changes and reduces tumorsphere and xenograft growth.
Why it matters: The study proposes a VRK1–SOX2 relationship that helps maintain a proliferative, progenitor-like tumor state and may be targetable through differentiation-based strategies.
Why for Yiru: This complements your interests in single-cell states and tumor plasticity by connecting a kinase dependency with cellular immaturity and self-renewal.