NewScience is the AI research agent that helps you identify high-quality hypotheses through built-in benchmarks, reproduce all the important metrics, and let you build meaningful science faster.
AI systems propose molecules, predict structures, and generate benchmark claims faster than any review process can keep up. Unverified machine output doesn’t reduce noise — it multiplies it.
A headline claim is published without the parameters needed to reproduce it, and almost never independently re-executed. The reproducibility crisis is a verification gap at the level of the single claim.
Reputation still rests on citations, journal prestige, and recency — proxies for attention, not validity. Capital follows the most confident narrative rather than the soundest evidence, and a wrong read is expensive.
NewScience is an AI research copilot for biology — with retrieval, analysis, and a growing set of integrations in one environment. What sets it apart is a built-in verification engine that re-executes computational claims, so your pipeline runs on ground truth, not trust.
Finds, reads, and synthesizes the literature for you. Drop an article in for an assessment, or ask a question and get the 5–10 strongest results — opened every day.
Ranks that work by reproducibility and executability — the deterministic E-class and NSE score, not citation. The signal that makes the results trustworthy.
Physically re-executes a computational claim through dry-lab connectors and scores how closely it reproduces — signing every run with cryptographic provenance.
These four domains are where we focus today — each chosen because its claims can be physically regenerated by our verification layer. As the engine matures, we expand into more.
Docking papers claim a ligand binds a target with a specific energy — but rarely ship the files to prove it. We extract the parameters, re-dock the reported pairs, and score the gap between what was claimed and what actually reproduces.
The result: every docking claim you rely on gets a verified, traceable binding-energy comparison instead of a number you have to take on faith.
Structure predictions circulate as fact before anyone re-runs them. We re-execute folding claims through open models and measure how closely the regenerated structure matches what a paper reported.
Instead of trusting a static figure, you get a re-run with a scored structural comparison — turning “the paper says it folds this way” into evidence you can verify.
MD simulations are long, expensive, and almost never re-run by reviewers. We regenerate the trajectory from published parameters and compare energy profiles, conformational states, and stability claims against what was reported.
The longer and costlier the original run, the more valuable an independent reproduction becomes — and the harder it is to fake.
Genomics and next-generation sequencing pipelines produce computational claims at scale — variant calls, expression quantifications, assembly metrics. We extend the same regeneration discipline here: re-run the pipeline, compare the output, score the gap.
Datasets and bio-models become standalone, verifiable artifacts — not just files attached to a paper.
Connectors bring public databases, literature sources, and chemical identity services into one environment — so every claim traces back to an official source.
Coming soon
