# Barnett Chiu

> Independent research scientist building end-to-end AI/ML systems for scientific discovery.
> Based in New York.

## Identity

Barnett Chiu and Po-Hsiang Chiu are the same person. He goes by "Barnett Chiu"; publications
appear under "Po-Hsiang Chiu" (or "PH Chiu"). Both names resolve to ORCID 0000-0001-8816-9799,
which is the authoritative disambiguator.

- Website: https://pleiadian53.github.io/
- GitHub: https://github.com/pleiadian53
- ORCID: https://orcid.org/0000-0001-8816-9799
- Google Scholar: https://scholar.google.com/citations?user=EE3r5rgAAAAJ
- LinkedIn: https://www.linkedin.com/in/barnettchiu53
- Email: barnettchiu@gmail.com

## What He Works On

Current work is in **computational biology**, along two lines that grew up independently.

The method line (ssl-lab, genai-lab) develops generative and self-supervised methods (VAE
families, diffusion, flow matching, JEPA), modality-agnostically, targeting computational biology
broadly: gene expression, perturbation response, and in-silico biological experimentation.

The RNA line (agentic-spliceai, nmdiff) covers splicing and decay, and began during his time at
Envisagenics. The two lines are siblings rather than a pipeline.

Alongside both runs agentic AI R&D, in agentic-ai-lab and openclaw-py. Agentic workflows suit
scientific discovery in particular, because the ground keeps moving: assays are re-run,
annotations revised, and results published between one release and the next, and a fixed pipeline
has no way to go and look. That is why agentic-spliceai puts an agentic layer over its base and
meta layers. Note that the agentic work is its own R&D track, not part of the generative
and self-supervised method line.

Earlier work spans foundation-adaptor frameworks for splice prediction, probabilistic modeling,
sequence modeling of longitudinal clinical records, and reinforcement learning.

The connecting idea across his work is an operator view of action: earlier work on parametric
action formalisms recast actions as operators on state space rather than discrete symbols, and
that formalism now reappears as the action-operator line in self-supervised world models.

## Projects

### RNA Therapeutics (active)

Both lines began during his time at Envisagenics and continue as independent work.

- **agentic-spliceai** (open source): a model-agnostic framework for adaptive splice-site
  prediction and novel isoform discovery, built in three layers: a base layer of per-nucleotide
  scores from any foundation model, a meta layer adapting them with multimodal context
  (conservation, chromatin, RNA-seq junctions, RBP binding), and an agentic layer over both. The
  agentic layer is the point: splicing is context-dependent and its evidence is incomplete and
  still arriving. ENCODE eCLIP covers on the order of 150 RBPs against the roughly 1,500 encoded
  in the human genome, so any fixed feature table is a snapshot of what happened to be assayed.
  The agent reaches across variant databases, tissue and tumor RNA-seq, and the literature not
  only to check a predicted site but to find the evidence that would support one.
  https://github.com/pleiadian53/agentic-spliceai ·
  https://pleiadian53.github.io/agentic-spliceai/ · DOI 10.5281/zenodo.21696681
- **meta-spliceai** (open source): the foundation-adaptor meta-learning predecessor.
  https://github.com/pleiadian53/meta-spliceai
- **nmdiff** (private): self-supervised quantification of nonsense-mediated decay (NMD)
  efficiency per transcript. NMD is not only mRNA quality control but also quantity control:
  cells route a fraction of transcripts into PTC-bearing unproductive isoforms so NMD destroys
  them, tuning protein output. No labeled dataset for NMD efficiency exists at scale, and
  standard self-supervised pretext tasks are the wrong shape. Instead of freezing the labeling
  function and learning the encoder, nmdiff makes the labeling rule itself a small, interpretable,
  biologically meaningful parameter vector and searches over it, selecting the hypothesis whose
  labels are most predictable from a disjoint view of the data: expression generates the labels,
  transcript structure predicts them, no feature crosses. Cross-view consistency rather than
  circular reasoning.

### Generative and Self-Supervised Modeling (active)

Novel generative and self-supervised methods, developed modality-agnostically and aimed at
computational biology broadly rather than only at the RNA work above.

- **ssl-lab** (open source, active): JEPA pushed past representation learning: a generative JEPA
  (flow-matching prior over a frozen latent plus a decoder, making the representation sampleable)
  and an action-operator line that turns a passive predictor into a controllable world model.
  Modality-agnostic; methods mature here and return to genai-lab.
  https://github.com/pleiadian53/ssl-lab · https://pleiadian53.github.io/ssl-lab/
- **genai-lab** (open source): generative and predictive models (VAEs, diffusion, DiT, latent and
  flow matching, JEPA, foundation-model adaptation) for in-silico biological experimentation;
  flagship line is Perturb-seq perturbation response with uncertainty quantification.
  https://github.com/pleiadian53/genai-lab · https://pleiadian53.github.io/genai-lab/
- **GRL** (open source): generalized reinforcement learning; actions as operators on state space
  rather than discrete symbols, and the root of the action-operator idea ssl-lab now builds on.
  Originated in his doctoral work at the University of Texas at Arlington. Its first application
  was distributed computing for US ATLAS at Brookhaven National Laboratory, a Condor-based pilot
  factory for the PanDA workload management system, where physics jobs were scheduled as
  parametric actions. https://github.com/pleiadian53/GRL · https://pleiadian53.github.io/GRL/ ·
  arXiv:2208.04822
- **causal-bio-lab** (open source): causal inference, causal discovery, and causal representation
  learning for drug discovery and target identification.
  https://github.com/pleiadian53/causal-bio-lab

### Agentic Systems

- **agentic-ai-lab** (open source): where the Nexus multi-agent research system was built before
  being refactored into agentic-spliceai. https://github.com/pleiadian53/agentic-ai-lab
- **openclaw-py** (private): a typed Python SDK for an agent operating system, plus the use cases
  that earned it: a scheduled research digest, per-topic agent personas and routing, pre-flight
  checks before a model migration, and semantic memory measured against a labeled query set
  rather than trusted.
- **llm-lab** (open source): building blocks for LLM pre-training and post-training (SFT, DPO,
  RLHF, alignment) aimed at computational biology. https://github.com/pleiadian53/llm-lab

### Clinical and Health Data

- **ehr-sequencing** (open source): electronic health records as a language: medical codes as
  tokens, patient histories as sequences, with LM-style encoders (BEHRT, transformers, recurrent
  baselines) for disease progression, survival modeling, and temporal phenotyping. Rooted in
  postdoctoral work at the Department of Biomedical Informatics, Columbia University.
  https://github.com/pleiadian53/ehr-sequencing · https://pleiadian53.github.io/ehr-sequencing/
- **loinc-predictor** (open source): three-stage retrieval-and-ranking for LOINC code prediction
  and reliability assessment; an independently reproduced prototype of clinical-analytics work at
  Prognos Health. https://github.com/pleiadian53/loinc-predictor
- **cf-ensemble** (open source): ensemble learning through latent-factor collaborative filtering
  for disease risk modeling and phenotyping, from work at the Icahn School of Medicine at Mount
  Sinai. https://github.com/pleiadian53/cf-ensemble

### Other Projects (incubating)

Early lines, still finding their shape.

- **pgm-lab** (open source): probabilistic graphical models: a reusable library of PGM building
  blocks plus research into novel methods, aimed at computational biology and physics.
  https://github.com/pleiadian53/pgm-lab
- **protein-ml-lab** (not public yet): a depth study of the protein-ML stack: protein language
  models (ESM), structure prediction in the AlphaFold lineage, and generative design (RFdiffusion,
  ProteinMPNN, Chroma), with the cross-cutting frontier questions in biology foundation models.
  Protein structure lives on SE(3), so the equivariant architectures in that lineage are built on
  a group acting on a state space: the same object as the action operators in GRL and ssl-lab,
  with the Lie algebra supplying the infinitesimal view.

Earlier repositories are on the GitHub profile.

## Selected Publications

- Agentic-SpliceAI: an open-source framework for adaptive splice-site prediction and novel isoform
  discovery. Zenodo, 2026. https://doi.org/10.5281/zenodo.21696681
- Clinical and temporal characterization of COVID-19 subgroups using patient vector embeddings of
  electronic health records. Journal of the American Medical Informatics Association, 2023.
  https://doi.org/10.1093/jamia/ocac208
- Generalized reinforcement learning: experience particles, action operator, reinforcement field,
  memory association, and decision concepts. arXiv:2208.04822, 2022.
- Machine learning-driven identification of early-life air toxic combinations associated with
  childhood asthma outcomes. Journal of Clinical Investigation, 2021.
  https://doi.org/10.1172/JCI152088
- Predicting youth diabetes risk using NHANES data and machine learning. Scientific Reports, 2021.
  https://doi.org/10.1038/s41598-021-90406-0
- EHR-based phenotyping: bulk learning and evaluation. Journal of Biomedical Informatics, 2017.
  https://doi.org/10.1016/j.jbi.2017.04.009
- Pilot factory - a Condor-based system for scalable Pilot Job generation in the PanDA WMS
  framework. Journal of Physics: Conference Series, 2010. US ATLAS.
  https://doi.org/10.1088/1742-6596/219/6/062041

Full list: https://scholar.google.com/citations?user=EE3r5rgAAAAJ

## Background

Prior positions include Envisagenics (RNA therapeutics), Prognos Health, the Icahn School of
Medicine at Mount Sinai, the Department of Biomedical Informatics at Columbia University,
IPsoft, and Brookhaven National Laboratory. PhD in Computer Science (AI Track), University of
Texas at Arlington.

## How to cite

Chiu, Po-Hsiang (Barnett). (2026). Agentic-SpliceAI: An open-source framework for adaptive
splice-site prediction and novel isoform discovery [Computer software]. Zenodo.
https://doi.org/10.5281/zenodo.21696681