Shucheng Cao

Quantitative inference / computational systems

Shucheng Cao 曹书诚

Also known and published as Bangli Cao

I build quantitative systems that make difficult evidence easier to trust.

I am a Quantitative Life Sciences PhD candidate at McGill University. I build end-to-end systems across causal inference, machine learning, simulation, and auditable AI, with current applications in genetics and biomedicine.

Email Google Scholar ORCID GitHub LinkedIn CV
Portrait of Shucheng Cao
PhD candidate
McGill University
Montréal, Canada

01 News

News

New writing, research milestones, talks, and selected presentations.

  1. Upcoming Poster accepted for ASHG 2026 in Montréal
  2. Blog Two agents, and no way to tell if either one workedwhy I helped start a benchmark instead of building a third
  3. Blog Building an AI agent that can’t fake drug-target evidencefrom a metabolite in blood, through the gene behind it, to whether it is worth a drug
  4. Blog Building an AI agent that can’t fake drug-target evidencefrom a protein to whether it is worth a drug
  5. Presentation Presented CausalSentinel: An Agent That Has to Show Its Receipts at the CABS final presentation.
  6. Lab talk Presented a preliminary two-step Mendelian-randomization workflow on adiposity → circulating metabolites → MASLD.
  7. Paper Our study separating osteoarthritis-specific genetic effects from obesity was accepted at Nature Communications
  8. Poster Presented Integrative proteogenomic Mendelian randomization identifies and refines obesity-driven circulating mediators of MASLD at ESHG 2026 in Gothenburg (P17.030.F).
Earlier news and presentations
  1. Lab talk Presented a journal discussion on a genetic map of human metabolism across the allele-frequency spectrum.
  2. Lab talk Led a journal discussion on the genome-wide characterization of circulating metabolic biomarkers.
  3. Research talk Presented a two-step Mendelian-randomization proposal on metabolite mediators of liver cirrhosis.
  4. Lab talk Led a journal discussion on a multi-omic definition of metabolic obesity.
  5. Talk Gave an oral presentation, Causal Inference of Obesity-Related Pathways in Metabolic Dysfunction-associated Steatotic Liver Diseases (MASLD), at McGill QLS Research Day.
  6. Invited talk Gave an invited talk on obesity-related pathways in MASLD at the Cardiovascular & Metabolic Diseases Innovation Center, Guangdong Medical University.
  7. Invited talk Presented Impact of Obesity-Related Proteins in Metabolic Dysfunction-associated Steatotic Liver Disease (MASLD) at the Lady Davis Institute Epidemiology Round Table.
  8. Service Coordinated LDI Research Day, bringing together 300+ attendees from 10+ institutions.
  9. Award Awarded the FRQNT Doctoral Scholarship (CA$25,000 per year).
  10. Poster Presented Impact of BMI-driven proteins on nonalcoholic fatty liver disease (NAFLD) at ASHG 2024 in Denver (Board 5087F).
  11. Award Won Best Product–Market Fit at the AWS Retail GenAI Hackathon.

02 Selected work

Selected work

Different domains, one recurring job: find where an apparently strong result can fail, build the checks into the system, and leave evidence another person can inspect.

01 Causal inference
McGill

Causal targets for metabolic liver disease

Protein–disease associations can disappear under confounding or a change in measurement platform. I built an R/HPC pipeline that combines Mendelian randomization, colocalization, multiple-testing control, and independent replication to move from thousands of candidates to four high-confidence targets.

  • 790,000+ participants
  • 2 proteomic platforms
  • 4 replicated targets
First-author MASLD manuscript submitted Related causal-inference paper

02 Auditable AI
CABS 2026

Evidence-grounded AI for drug-target prioritization

LLMs can sound certain even when their own sources do not support the conclusion. I led the engineering of OpenCausal, the causal-evidence component of CausalSentinel: a Python/Streamlit system that retrieves public biomedical evidence, records provenance, renders reports deterministically, and rejects unsupported model-written claims before release.

  • 8 public databases
  • 62 regression tests
  • 991 evidence dossiers
  • 101,543 estimates
Code and methods Card viewer Blog case study

03 Geometric ML
McGill

A reproducible GNN benchmark for neuron classification

To test whether a training-free tropical-geometry method could beat a strong deep-learning baseline, I reproduced MorphoGNN and built a deterministic PyTorch benchmark across three datasets. The audit found a silent mapping bug affecting 97% of samples; after leakage checks and five seeds, the corrected result became defensible.

  • 3 datasets
  • 97% mapping error found
  • 80.1 ± 1.0% balanced accuracy
AAAI manuscript submitted · co-author

04 Model evaluation
IEEE TPAMI

Can medical AI generalize across hospitals?

I helped build AbdomenCT-1K and used cross-hospital evaluation to measure how leading multi-organ segmentation models degrade when scanners, protocols, and patient populations change. The study made deployment risk measurable rather than assumed.

  • 1,112 CT scans
  • 12 hospitals
  • 4,000+ teams using the data
Paper Dataset and toolkit

05 Molecular simulation
KAUST

Simulation-guided design for an aggregation inhibitor

I ran controlled molecular-dynamics experiments that changed hydrophobicity and aromaticity separately, then turned the resulting mechanism into a quantitative ranking of new peptide designs. The work challenged a two-decade field assumption and produced a first proof-of-concept design for blocking amyloid aggregation.

  • Controlled MD experiments
  • Top-ten KAUST MSc thesis
MSc thesis

A working rule: traceability is part of the result. Another person should be able to reconstruct the inputs, assumptions, checks, and failure modes behind it.

03 Publications

Selected publications

Peer-reviewed and accepted papers behind the work above, ordered from most recent.

  1. 2026 Disentangling osteoarthritis-specific genetic effects from obesity to identify novel therapeutic targets C.-Y. Su, M. Hasebe, D. Tan, Bangli Cao, et al. Nature Communications — accepted / in press; preprint available. Related work: causal inference ↑
  2. 2023 A PDA-Functionalized 3D Lung Scaffold Bioplatform to Construct Complicated Breast Tumor Microenvironment for Anticancer Drug Screening and Immunotherapy W. Zhang, Y. Chen, M. Li, S. Cao, et al. Advanced Science.
  3. 2022 An improved cell line-derived xenograft humanized mouse model for evaluation of PD-1/PD-L1 blocker BMS202-induced immune responses in colorectal cancer Y. Zhang, J. Zhang, S. Cao, et al. Acta Biochimica et Biophysica Sinica.
  4. 2022 AbdomenCT-1K: Is Abdominal Organ Segmentation a Solved Problem? J. Ma, …, S. Cao, et al. IEEE Transactions on Pattern Analysis and Machine Intelligence. Related work: out-of-distribution evaluation ↑

Thesis

See the full publication record on Google Scholar ↗

04 Blog

Blog

Longer accounts of how a result failed, what the failure exposed, and which checks were worth keeping.

Browse all Blog posts →

05 Contact

Get in touch

I am interested in quantitative research, applied AI, and research-engineering roles wherever evidence quality and model reliability matter. Email is the best way to reach me.

shucheng.cao@mail.mcgill.ca