Hackathon 2026 · Submissions
What teams built
An archive of the projects teams shipped over the three days — most went from idea to a working prototype, built with Claude Code. Where a team shared a repo or site, the card links out; the rest presented from slides. A gold 🏆 badge marks the category winners, listed first.
PeerReviewAgents
A multi-agent system that simulates a full journal editorial board on any PDF manuscript: eight specialist reviewers each cover a different dimension, an advocate-vs-skeptic debate and a simulated author rebuttal feed an Editor-in-Chief verdict (accept / minor / major / reject) plus journal-fit picks. Runs on Claude via Amazon Bedrock AgentCore, with live arXiv search to ground novelty checks. In validation it matched the real decisions on three rejected papers and three judges' accepted manuscripts.
Patrick Garrett, Ricard Garcia, Yifan Wang, Rama Aldakhlallah, Aleix Navarro
View repo 🏆 Most Innovative Electrophysiology · visualizationBrain Organoid Builder (BOB)
Reproduces neuronal firing-sequence analysis from microelectrode-array recordings of brain organoids and renders it as a spatial 3D animation. Pulls a human-organoid recording from DANDI, spike-sorts with Kilosort4, and ships a Blender add-on that animates each unit's firing rate over its location on the array.
Arjun Pamidi, Anyang Chen — Diercks Lab
View repoCryo-EM 2D-class junk classifier
Removes "junk" 2D class averages from negative-stain particle stacks to raise experiment throughput. Compared a Claude-rebuilt ResNet, a from-scratch Claude pipeline (DINOv2 / ResNet-18) and a VLM agent against human and CryoSift baselines — the from-scratch single-shot build came closest to human classification.
Charles Bowman, Daniel Montiel, Nicholas Lee, Andy Tran, Meghna Chandrasekar
LLM cell-type annotation benchmark
Benchmarks LLMs against reference mapping for single-cell RNA-seq cell-type annotation. Compares zero-shot marker-gene labeling — with and without tissue plus PubMed context — against CellTypist across PBMC and pancreas, scoring accuracy, runtime and memory. Context-grounded LLMs matched or beat reference mapping on the less-canonical tissue.
Chenhao Wu, Eduard Ansaldo, Juyeon Park, Xing Zhang
View repo 🏆 Best Presentation Visualization & AI agentsMol*-MCP — Talk to your structure
A thin MCP layer that drives the web-based Mol* viewer from plain English, turning a flat list of sequence or variant hits into live 3D structure with no installs. Natural-language intent maps to viewer verbs — load, highlight residues, color by, measure, spin. Demo: painting 27 SARS-CoV-2 RBD escape positions onto PDB 6W41.
Renhao Luo, Meng Yuan
View repo 🏆 Cross-Disciplinary Education · visualizationViralBeat Studio
A browser app for exploring icosahedral symmetry in virus capsids by mapping music to structure. Pick a virus, load an audio track, assign musical elements to symmetry axes, and watch an audio-reactive Mol* animation — a multisensory take on molecular-symmetry teaching.
Autin Lab
View repo 🏆 Claude's Choice NMR · deep learningSpinhance
Recovers field-independent spin-system parameters — chemical shifts, J-couplings, proton counts — from the overlapping peaks of 90 MHz benchtop ¹H NMR, so a blurry spectrum can be re-simulated at any field. A transformer encoder–decoder trained on 3.13M synthetic spin systems, paired with a differentiable quantum spin-Hamiltonian simulator and a spectral-consistency loss.
Lucas Abounader, Sam Mansfield, Yiming Zhang — Shenvi & Seiple Labs
View site Cryo-ET · machine learningTomoAnnotator
Weakly-supervised organelle detection in cryo-ET tomograms: paint a few brush strokes per class on one or two volumes, train a compact 2D / 2.5D / 3D CNN in minutes, then slide it over new tomograms for per-class presence/absence calls. Runs end-to-end on real Grotjahn Lab data, with SLURM, EC2 and S3 wiring and a Claude-written summary.
Michaela Medina, Hamidreza Rahmani — Grotjahn Lab
View repoAgentic AI for digital biomarker discovery
An orchestrator and parallel subagents run a high-throughput hypothesize → extract → train → evaluate loop over NHANES accelerometry, behind a data gate with leakage bans and an independent Critic reviewer. Explored 13 hypotheses across 9 outcomes; activity volume added a material +0.0084 outer-test AUC to 10-year mortality prediction.
Heguang Lin, Michael Ko, Giulia Milan — Quer Lab
Predicting drug safety from proteomics
A closed-loop compound-safety workflow built solo in a day. A scaffold-aware screen recovered 28 viable analogues from 419 where a naive filter kept only 6, and a 14-module proteomics pipeline runs QC → normalization → FDR statistics → dose–response → ranking, with Claude-API confidence scoring.
Casimir Bamberger — Scripps Research
PQBP1 as a conformational sensor for HIV-1 capsid
A Claude-assisted integrative-modeling pipeline placing PQBP1 (1–265) against the HIV-1 capsid hexamer, fusing XL-MS, HDX-MS, NMR and docking (MODELLER + HADDOCK) into a three-state autoinhibited → capsid-bound → cGAS-competent mechanism. About 376 structural models generated overnight on Scripps HPC.
Dale Allen — Chanda Lab
Protein property prediction
Frozen protein-language-model embeddings feeding a lightweight prediction head to estimate melting temperature (Tm) on the FLIP meltome split. Benchmarked ESM2 against ESMC embeddings across kNN, ridge, MLP and attention heads, reaching ~0.72 correlation.
Anastasiya Kuznetsova (Balch Lab), Qishan Liang (Miller Lab)
QuantAb
Ran antibody binding-affinity predictions on quantum computers, accessed through AWS, and found the quantum approach outperformed the classical baseline — promising enough that the team plans to keep exploring it.
MEMpro
Predicted protein–protein interactions with machine-learning models and benchmarked them against the team's own physics-based calculations, running many models at scale on AWS.
Siglec-6 binder
Docking analysis to surface small molecules predicted to bind Siglec-6, with wet-lab evaluation planned as a follow-up. Docking ran on AWS EC2 with data stored in S3.
About
A low-stakes sandbox for high-stakes ideas.
A hands-on hackathon where researchers across Scripps come together to dig into real biological datasets, experiment with new tools, and build something together. No prior experience required, just curiosity and a willingness to dig in.
Bring your own project, or pick from the suggested ideas — either path is welcome. Same goes for teams: come with a team already assembled, or form one on the day during Friday morning’s project pitches and team-formation session.
Who can participate
Anyone at Scripps who’s curious — graduate students, postdocs, staff scientists, and researchers at any level of computational experience. If you’ve never written a line of code, that’s okay. Domain expertise, biological intuition, and fresh perspectives are just as valuable as technical chops, and projects are scoped so that wet-lab biologists, computational researchers, and engineers can contribute meaningfully on the same team.
This is an internal Scripps Research event. Participation is limited to Scripps Research personnel — faculty, postdocs, students, and staff. External collaborators are not eligible to join.
Within Scripps, anyone is welcome: wet-lab biologists, computational researchers, software engineers, ML practitioners, and data scientists. Participants do not need to be experts in everything — projects are scoped so contributors from different backgrounds can work meaningfully on the same team.
Format
The main hacking day is on Friday, with team presentations and awards the following Monday:
The in-person component runs on Friday and Monday only. The weekend in between is unstructured and entirely optional — there is no designated venue, and teams can choose whether to work, how much, and from where.
- Friday, May 29 (in person) — kickoff, keynote, project pitches, team formation, training sessions (agentic AI bootcamp, AWS 101), and the first hacking session.
- Weekend (May 30–31) (optional, remote) — open hacking time for teams that want it. No designated venue, no required sessions.
- Monday, June 1 (in person) — project presentations, judging, prizes, and wrap-up.
See the agenda for timings.
What you’ll get out of it
- A small, focused team built around a problem you actually want to solve.
- Hands-on time with tools you may not otherwise have access to — enterprise agentic AI tooling and specially provisioned AWS compute and storage.
- Feedback from organizers and peers throughout the event.
- New skills and new collaborators to bring back to your own research.
- A finished project page on this site, archiving what your team built.
Topics
Themes & project areas
Project proposals can fall into any of the following broad themes — but we encourage cross-cutting work.
- Protein structure & design — folding, docking, generative design, binders, antibody engineering.
- Single-cell & spatial omics — atlas integration, foundation-model embeddings, lineage tracing, perturbation analysis.
- Machine learning & deep learning for biology — pretrained models, benchmarking, interpretability, active learning.
- Genomics & population genetics — variant interpretation, GWAS, ancestry, simulation.
- Imaging — segmentation, registration, hyperspectral analysis, interactive viewers.
- Tooling & infrastructure — pipelines, WebAssembly, GPU acceleration, reproducible environments, web portals for shared tools.
- Open data & FAIR practices — metadata, schema design, ontology alignment, archival.
Submit a project proposal
If you want to lead a project, submit a proposal by the deadline. Proposals should describe:
- The problem you want to tackle and why it matters.
- The data, tools, or models the project will build on.
- What a successful three-day output looks like.
- The skills or roles you’re hoping to recruit on your team.
Schedule
Agenda
Day 1 — Friday, May 29
Kresge Library
| Time | Session |
|---|---|
| 8:30 am | Coffee |
| 9:05 am | Welcome |
| 9:15 am | Keynote — Jamie Williamson |
| 9:30 am | Project pitches |
| 9:40 am | Team formation |
| 10:00 am | Project outlining |
| 10:15 am | Agentic AI bootcamp (Dr. Benjamin Good) — Claude Code setup, example use-cases, limitations, Scripps HPC skill |
| 11:00 am | AWS 101 (Omar Tabbakha, Nadeem Bulsara, Dr. Krutika Khinvasara) — getting familiar with working within AWS, Amazon Open Data, Amazon Bio Discovery, Amazon Bedrock |
| 12:15 pm | Lunch (hackathon & training-session participants) |
| 12:45 pm | Time for hacking |
| 4:45 pm | End-of-day announcements |
| 5:00 pm | Dinner (hackathon participants only) |
Day 2 — Monday, June 1
Kresge Library
| Time | Session |
|---|---|
| 10:00 am | Welcome |
| 10:10 am | Project presentations (5–7 min/project) |
| 12:00 pm | Lunch / judge deliberations |
| 12:30 pm | Prize announcements |
| 12:35 pm | Wrap-up and acknowledgements |
Suggested
Project ideas
A menu of suggested projects to spark ideas — these are starting points teams can pick up, not the finished work. For what teams actually built, see What teams built →. Click any idea for its goals, methods, and resources. Want to add one? Submit a project idea →
Benchmarking LLMs for automated scRNA-seq analysis
Test which LLM produces the fastest, most accurate cell-type annotation pipeline using expert-curated public scRNA-seq datasets as ground truth.
View idea Structural biology & visualizationCaspar–Klug T-Number Explorer
An interactive tool for exploring the geometry of viral icosahedral capsids — T-number, triangular subdivisions, pentamers and hexamers, and how simple rules generate complex viral shells.
Lead: ,
View idea Neuroscience, machine learning, computational modelingInferring brain organoid structure from electrode recordings
Reverse-infer the structure of a brain organoid from multi-electrode recordings using computational models and machine learning.
View idea Visualization & AI agentsNatural-Language Control for Molstar
Drive the Molstar molecular viewer with chat — generate MolViewSpec scenes (and, as a stretch, MolViewStories) from plain-English requests.
Lead: ,
View idea Structural biology & outreachPDB to 3D Print
One-click conversion from any PDB ID or uploaded structure file to a printable, mesh-repaired 3D file — with a short scientific 'structure card' to go with it.
Lead: ,
View idea Integrative structural modelling, innate immunityPQBP1 as a central regulator of inflammation in neurodegeneration
Use integrative modelling (HADDOCK) to build structural hypotheses for how PQBP1 — a small disordered adapter — recognises oligomeric ligands (HIV-1 capsid, Tau fibrils, α-syn fibrils) and recruits cGAS to drive inflammation.
Lead: Dale Allen, Chanda Lab, Scripps Research
View idea ImmunologyPeptide–MHC binding affinity vs. viral evolution in COVID-19
Test whether peptide–MHC binding affinity correlates with viral mutation rates in SARS-CoV-2 across the pandemic.
View idea Protein engineering, machine learningProtein language model embeddings for property prediction
Extract embeddings from protein language models, train prediction heads on a chosen property, and benchmark across PLMs and head architectures.
View idea Research tooling & AI agentsProtocol-to-Agent Builder
Turn any protocol — wet-lab SOP, microscopy workflow, computational pipeline — into a grounded, queryable assistant that can answer practical execution questions and produce checklists.
Lead: ,
View idea Virtual screening, immunologySiglec-6 small-molecule binding
Computationally screen and dock candidate analogues against Siglec-6 to improve on the lab's existing weak (~75 µM) lead — promising hits get synthesised and validated wet-lab (SPR for K_d, CETSA in live cells).
Lead: Julien Lee Heberling, Huang Lab, Scripps Research
View idea Behavioral analysis, machine learningTooling for C. elegans behavior analysis
Apply or build behavioral analysis tools to study how specific motor neurons shape C. elegans behavior, using a perturbation video dataset.
View idea Single-cell, transcriptomics, machine learningTraining a 'virtual cell' model for gene expression inference
Train a virtual-cell model from public scRNA-seq perturbation data and benchmark it against entries from the November 2025 Virtual Cell Challenge.
View idea AI agents, scientific toolingTrying and developing agentic AI for science
Stand up agentic AI scaffolds (Biomni, Claude Scientific Skills, Claude for Life Sciences) and either evaluate them or extend them on a real Scripps research workflow.
View idea Mixed reality & conversational AIVirtual Professor Art
Talk to a virtual scientific mentor in mixed reality — real-time conversation, synthesized voice, and an animated avatar that can react to what the headset sees.
Lead: ,
View idea Open callHave a project idea?
Pitch your own project — four required fields, takes a couple of minutes. Approved ideas land here as a card.
Submit a project ideaMaterials
Resources
A growing collection of materials for hackathon participants — training sessions, walkthroughs, and slides from event presenters.
Other resources
- Scripps Hackathon Starter — A ready-to-fork template repo. Comes with a pre-configured Claude Code workspace (AWS and Garibaldi HPC skills), setup guides for both platforms, and three Hello World exercises that validate your AWS SSO and HPC accounts before the event kicks off.
- Claude Chat, Cowork and Code, from setup to tips and tricks — Dr. Talmo Pereira (Salk). A walkthrough of Claude Code from installation through advanced workflows. Recording · Notes.
- Life Science AI Ecosystem — Curated, daily-refreshed catalog of Claude AI components (skills, MCP servers, plugins, connectors) applicable to life-sciences research, with recipes pairing research problems to recommended tool combinations.
Event materials
Slides from event presenters will be posted here as they’re shared.
Organizers
Organizing committee
The team running the hackathon.
Ian Newman
Gabriel Ong
James Caviness
Rotimi Omorodion
Quentin Tallon
Shelby Ferrier
Richard Gast
Lisa Janssen
Judges
Judges
The panel evaluating final project presentations on Monday.
Andrew Su
Scripps Research
Stefano Forli
Scripps Research
Megan Ken
Scripps Research
Prizes
Prizes
Prizes are announced on Monday at 12:30 pm following judge deliberations over lunch. Seven prizes are awarded:
- Grand Prize — the team with the highest combined score across all five rubric categories.
- Most Innovative — for the most original approach or genuinely new technique.
- Best Scientific Merit — for the project tackling the most significant scientific problem.
- Biggest Impact — for the project most likely to change practice or unlock follow-on research.
- Best Presentation — for the team that communicated their work most clearly and compellingly.
- Cross-Disciplinary Award — for the project that most meaningfully integrated multiple disciplines.
- Claude’s Choice Award — selected by the AI judge.
Presentation format & eligibility
- Each team gives a 5–7 minute presentation on Monday morning.
- Submit your slides on Monday to be eligible for prizes.
See the judging rubric for how projects are scored.
Rubrics
Judging rubric
Projects are scored by three faculty judges and one AI judge (Claude) across five categories worth 10 points each — 50 points total.
Scoring scale (per category)
| Score | Descriptor |
|---|---|
| 9–10 | Exceptional — exceeds expectations in a meaningful way |
| 7–8 | Strong — clearly meets the standard with notable strengths |
| 5–6 | Adequate — meets the basic standard with some gaps |
| 3–4 | Developing — partial effort, key elements missing or weak |
| 1–2 | Insufficient — minimal engagement with the criterion |
| 0 | Not addressed |
Categories
- Innovation & Creativity — fresh perspectives, novel methodology, or a creative solution to a real data-processing problem. Did the team take any intellectual risks?
- Scientific Merit — the significance of the problem and how meaningfully the solution advances it. The importance of the science, not the technical execution.
- Technical Quality & Execution — methodological soundness, reproducibility, and how accessible the outputs (and code, if any) are to others.
- Communication — clarity and effectiveness of the team’s presentation, both verbally and visually, for specialists and non-specialists alike.
- Interdisciplinary Approach — meaningful integration of two or more disciplines (biology, computer science, statistics, clinical science, chemistry, physics, etc.).
Partners
Partners & sponsors
We’re grateful for the support of the following institutes and organizations.
Interested in sponsoring or partnering on the next hackathon? Get in touch with any of the organizers.
FAQ
Frequently asked questions
How will instructions and updates be communicated?
Through the CBB Hackathon Slack workspace. All instructions, schedule updates, training-session links, and announcements will be posted there before and during the event. It’s also the fastest way to reach the organizers if you have a question or run into a problem.
Who can participate?
This is an internal Scripps Research event. Participation is limited to Scripps Research personnel — faculty, postdocs, students, and staff. External collaborators are not eligible. Within Scripps, wet-lab biologists, computational researchers, software engineers, ML practitioners, and data scientists are all welcome.
Where will the hackathon be held?
The in-person sessions on Friday and Monday are at Kresge Library on the Scripps Research La Jolla campus. Look for signage at the entrance on the day of the event. Parking is available on campus; allow extra time on Friday morning to find a spot and check in.
How will team formation work?
Both options work — come with your own team, or form one on the day. Friday morning starts with project pitches followed by a team-formation session, so you can pitch your own idea, join one that’s already on the board, or stick with a group you arrived with.
Are projects predetermined?
No. A handful of suggested projects are listed on the Projects section as starting points, but you are free to come up with your own. You can also remix or extend a suggested project — the pitch session on Friday is the place to propose whatever you want to build.
What if I need to step out for portions of the hackathon?
That’s completely fine. We know participants have meetings, experiments, and other obligations — come and go as you need. The weekend hacking is entirely optional and remote, so there’s no expectation to be present for any particular block of time outside the Friday and Monday sessions you want to attend.
Is there a registration fee?
No. Registration is free for eligible participants. Sign up on the event page.
What should I bring?
A laptop, your charger, and any specialty adapters you need. Power strips will be available at every team table.
Will food be provided?
Yes. Lunch is provided on both days, and dinner is provided on Day 1 for hackathon participants. Let us know any dietary needs on the registration form.
Can I attend remotely?
The Friday and Monday sessions are in-person at Kresge Library and not streamed. The weekend in between is optional and unstructured — teams that want to keep working can do so from anywhere.
Will there be GPUs and compute available?
Yes. Amazon Web Services is providing cloud compute, including GPU instances — the AWS 101 session on Day 1 walks through getting set up. The Scripps HPC environment is also available; both are covered in the Day 1 morning sessions.
Will I have access to Claude during the event?
Yes. Claude (including Claude Code) access will be provided. The Day 1 Agentic AI bootcamp walks through Claude Code setup, example use cases, limitations, and how to plug into the Scripps HPC skill.
Who owns the code we write?
Teams retain ownership of what they build. We encourage open-source releases under permissive licenses, but the choice is up to each team.
Will there be prizes?
Yes — final project presentations on Day 2 are followed by judge deliberations and prize announcements at lunchtime.
What happens after the hackathon?
Each team’s project gets a permanent page on this site documenting goals, methods, and outcomes. Teams are encouraged to keep going beyond the event.
What's the code of conduct?
Be respectful, be inclusive, and contribute to a space where everyone can do their best work. Issues can be reported to any organizer in person or via direct message on Slack.
Contact
Get in touch
The Slack workspace is the primary channel for hackathon communication. All instructions, updates, and announcements will be posted there, and it’s the fastest way to reach the organizers if you hit any problems.
Join the CBB Hackathon Slack →
You can also reach out to any of the organizers listed above directly.