Cross-resolution RNA benchmarks
How should models be judged when moving between bulk cohorts and single-cell atlases?
Independent non-commercial research collective
Virtual Human Lab
An open, independent research collective building RNA-modeling infrastructure, AI-enabled experiments, and public research memory for virtual-human biology.
What we are
A free research collective for the AI biology era.
Virtual Human Lab is an independent, non-commercial group of researchers working at the intersection of bioinformatics and machine learning. We are not a formal institute attached to a government agency, university, or company.
Our premise is simple: a new scientific field should be learnable from the ground up. With AI as a tutor, AI as an experimental partner, and GPU access as the shared instrument, motivated researchers can move faster than traditional training paths often allow.
Public research memory
Papers, notes, and open problems belong on the same shelf.
Virtual Human Lab will publish formal outputs and the reasoning around them: preprints, technical notes, negative results, reading maps, and implementation logs.
Formal outputs
Papers & preprints
Only public papers and preprints, with citations, code, data, and model links.
Working memoryResearch notes
Short technical posts on datasets, model design, evaluation, and biological interpretation.
Collaboration surfaceOpen problems
Questions where a collaborator, dataset, GPU path, or method critique can change the work.
Open problems
What we want to make legible next.
Biological interpretability for RNA embeddings
Which signals are pathway-level, which are batch artifacts, and which are clinically useful?
GPU-efficient training loops
How can independent researchers train meaningful models without institutional-scale compute?
Research programs
Foundation models for RNA-scale virtual humans.
Our current work focuses on learning cellular state, tissue-scale signals, and cross-resolution translation from RNA-seq data.
Input
Bulk cohorts + scRNA-seq atlases
Representation
RNA embeddings + cell-state maps
Evaluation
Pathways, perturbations, clinical signal
Output
Models, notes, code, preprints
Single-cell foundation model in scRNA-seq
Models that learn cell identity, perturbation response, and disease-state geometry from single-cell transcriptomes.
Bulk-cell foundation model in bulk RNA-seq
Representations for cohort-scale transcriptomes that preserve clinically useful biological structure.
RNA-LLM
A multimodal LLM direction that connects language reasoning with RNA-seq embeddings for interpretation and hypothesis generation.
Bulk RNA-seq to single-cell RNA-seq
Cross-resolution modeling that studies whether population-level RNA signals can be mapped into single-cell-like cellular programs.
Institute model
Free participation, serious work.
-
Self-directed entry
We welcome researchers who want to learn a new technical field without waiting for permission from a traditional lab hierarchy.
-
AI-native experimentation
We treat AI systems as learning infrastructure, coding partners, literature companions, and experimental accelerators.
-
GPU-centered collaboration
We are independent, but we actively welcome collaborations with institutions, universities, companies, and public organizations that can support GPU resources.
Members
A small starting group, built to stay open.
Current member backgrounds are listed for context only. Virtual Human Lab is an independent collective and does not claim institutional affiliation.
Collaborate
Bring a clinical question, dataset, or GPU path.
We welcome collaborations with researchers, labs, universities, companies, and public organizations interested in AI biology, transcriptomics, and virtual-human modeling.