Computational Screening vs Lab Experimentation: Which Comes First?

Screening comes first, and it isn't close. Computational screening evaluates thousands of candidate materials against every known constraint for roughly the cost of compute, while a single lab campaign consumes months and serious budget per candidate. The lab's job isn't exploration anymore — it's validation of the few candidates that survive screening.

Framing this as a rivalry misses the point. The lab is the only place a material becomes real. But treating the lab as your search engine — synthesizing your way through a design space one sample at a time — is how materials discovery ended up with development timelines measured in decades. The AIMS methodology exists because the two tools have different jobs, and putting them in the right order changes the economics of the whole field.

What can each approach actually do?

Read the last row twice. Screening is a search instrument; the lab is a proof instrument. Most wasted R&D money comes from using the proof instrument to do search.

Why does screening-first win?

Because elimination is cheap in silico and brutal in vitro. For most materials problems, the overwhelming majority of candidates fail — wrong stability window, toxic byproduct, unmanufacturable at target cost. Every one of those failures discovered in a lab costs a synthesis run. Every one discovered computationally costs approximately nothing. Constraint-satisfaction screening, the approach AIMS is built on, makes elimination the explicit goal: compile every testable physical, chemical, and economic constraint, then discard everything that violates any of them. What's left is a shortlist that has already survived every check the published literature makes possible.

Because screening includes the constraints labs discover too late. A classic lab-first failure: the material works beautifully and can't be manufactured at price, or degrades into something toxic. AIMS scores candidates against all constraints simultaneously — including cost, manufacturability, and destructive byproducts — at stage 4 of its pipeline, before anyone touches a beaker. Byproduct analysis at screening time is dramatically cheaper than byproduct discovery at scale-up time.

Because it changes what the first experiment is. When a candidate reaches the lab pre-ranked and pre-falsified, you don't need a multi-year program to investigate it. You need the minimum viable experiment — the cheapest, fastest measurement that can confirm or kill the hypothesis. That's stage 6 of the AIMS pipeline, and it's the hinge of the whole methodology: the lab's first contact with a candidate is a decisive test, not an open-ended exploration.

What can't screening do?

Be honest about this, because overclaiming is how AI-for-science gets a bad name.

• It can't prove anything. A screened candidate is a hypothesis with a survival record. Until it's synthesized and measured, it's not a material — it's a well-defended prediction.
• It inherits the literature's blind spots. Constraints come from published data. Where the literature is wrong or silent, screening can't see. This is why premise verification and citation auditing are mandatory stages, and why structured falsification attacks every top candidate before it's trusted.
• It doesn't capture synthesis surprises. Kinetics, defects, and processing quirks show up in the flask, not the model. The lab keeps its monopoly on that knowledge.

How does AIMS split the work?

AIMS runs the computational side end-to-end — premise verification, literature review, constraint compilation, simultaneous scoring, falsification, and MVE design — and publishes the results as papers with clear experimental pathways. To date that's produced 41 research papers and 78 filed provisional patents (~1,275 claims) across 40 scientific domains, from PFAS water purification to zeolite-confined superconductors.

The experimental side is deliberately a partnership model. Every AIMS discovery ships with its minimum viable experiment already designed, and the program actively seeks universities, national laboratories, and corporate R&D teams to run them. Computation finds and defends the candidates; institutional labs make them real. What that division of labor does to the budget is covered in what materials R&D actually costs, and the background on the search side is in what AI-driven materials discovery is.

FAQ

Can computational screening prove a material works?

No. Screening produces ranked, defensible hypotheses. Proof requires synthesis and measurement in a lab. What screening changes is which experiments are worth running — the lab validates a shortlist of survivors instead of exploring the whole space blind.

Why not just go straight to the lab?

Because lab experiments cost orders of magnitude more per candidate than computational evaluation, and the candidate space for most materials problems is enormous. Screening first means the expensive instrument — the lab — is pointed only at candidates that already satisfy every known physical, chemical, and economic constraint.

What is a minimum viable experiment (MVE)?

The cheapest, fastest experiment that can confirm or kill a hypothesis. It's stage 6 of the AIMS pipeline: instead of designing a multi-year research program, design the single decisive measurement. If the hypothesis dies, you've spent little; if it survives, you've earned the case for deeper investment.

Is AIMS research computational or experimental?

AIMS research is computational with clear experimental pathways. Every discovery ships with a designed minimum viable experiment, and AIMS actively seeks institutional partners — universities, national labs, corporate R&D teams — to carry candidates from computation into experimental validation.