Delta

Inspiration

Life-science VCs usually find a theme once a company exists, and by then the round is crowded and priced up. Every sourcing tool we looked at (Harmonic, PitchBook, CB Insights) starts at the company or the founder. But the signal shows up years earlier, in papers, grants and trials. We wanted to read it before anyone raises.

## What it does Delta ranks life-science research areas by how they're moving right now, and says whether each one is too early, on time or too late for an early-stage VC. It scans 1,423 OpenAlex life-science topics and zooms into specific niches, such as GLP-1 vs amylin or in vivo CAR-T. For each area it shows the evidence behind the call: papers by translation stage (mouse, primate, human), new NIH grants, newly funded labs, the first companies, and the scientists leading the field. Areas where the research is heating up but few companies exist yet are flagged as white space. Every gradebreaks down into the counts it came from, so a partner can check it in a minute.

How we built it

Data: OpenAlex (papers, authors, institutions, company affiliations), NIH RePORTER (new research grants and SBIR/STTR awards), ClinicalTrials.gov (industry sponsors), and akta.pro for recent company news per theme.
Point-in-time backtest: a feature at year T uses only data published by T. We tested on 34 past themes with known outcomes (17 successes, 12 fizzles, 5 open), each dated to the moment the money arrived.
Scoring: each signal is momentum over 3-year windows, z-scored across themes and averaged into a research grade and a company-presence grade.
Expert review: a domain expert on the team graded every theme's outcome label, and the shaky ones are left out of scoring.
Stack: Python pipeline, a static Next.js site on Vercel, Supabase.

Challenges we ran into

Raw paper growth is a trap. Themes that later fizzled, like sirtuin activators and senolytics, had more paper growth than themes that reached approval. We had to measure the kind of growth instead: work moving into primates and people, new NIH money, new labs getting funded.
Defining a theme by its query is fragile. ClinicalTrials.gov's search matched "Beam Therapeutics" to 316 radiation-beam trials until we searched company names in the sponsor field only.
Small numbers fake momentum. Going from 1 to 13 grants looked like a ×7 surge, so we damped momentum toward zero for tiny counts.
Labels are hard. DLL3 failed as an ADC and then succeeded as a T-cell engager. Expert review mattered more than any feature.

Accomplishments that we're proud of

A real backtest, not a demo on hand-picked examples. Research signals graded early: successes 0.05 vs fizzles −0.31. Company presence at the decision point: 0.18 vs −0.67.
Signals that were honestly negative stayed in the writeup, like raw paper momentum.
Every number on the site traces back to its source data.
The whole life-science scan cost about $4 of API credit.

What we learned

Volume of research doesn't tell a hit from hype; the direction of research does. Open data (OpenAlex, NIH RePORTER, ClinicalTrials.gov) is now good enough to do this for free. And a VC won't act on a score they can't take apart, so explainability is the product, not a feature.

What's next for Delta

Grow the list of fizzles, so the backtest can weight signals instead of averaging them.
Weekly alerts when an area moves into "on time".
A "founders to meet" list: the scientists in white-space areas who haven't started a company yet.
Patent and Chinese-literature coverage (OpenAlex covers about 37% of Chinese core journals).
Design-partner pilots with the funds we interviewed.