Science1 distinct publisher3 min readPublished
Its authors say an 18-dimensional property grid, which decides which chemical subspaces get docked at all, cuts compute by orders of magnitude, and they back the platform with nanomolar inhibitors of two targets plus co-crystal structures.
The Scientist · Science desk
Compiled by The ScientistSomething wrong?How this is made
The engineering worth studying here is subtractive. Docking all 69 billion compounds against a single target is a bill the paper itself names as a major bottleneck, one that worsens as vendor libraries push past a trillion molecules [13] [12]. AdaptiveFlow's answer is to make chemical space addressable: 18 molecular properties serve as the axes of a grid, cells of that grid are ranked, and only the promising subspaces get docked, with active learning available to sharpen the choice [3]. The reported saving is orders of magnitude of compute [3].
That design carries a cost the abstract does not price. A property grid assumes potency clusters in property-defined neighbourhoods. For enzyme active sites that is a reasonable bet. The paper's own motivation is the harder case, targets such as protein-protein interfaces and allosteric sites, where pockets are often shallow, dynamic or poorly defined [11]. The thing the abstract does not tell you is recall: of the actives an exhaustive dock of the full library would surface, how many survive prioritization.
Some arithmetic for scale sense. A typical high-throughput screen covers hundreds of thousands of compounds [9]; taking 500,000 as the figure, the dock-ready library is roughly 138,000 times larger [1]. Measured the other way, against on-demand chemistry that the authors put at up to 3 trillion molecules across vendors, 69 billion is about 2.3 percent [2] [12]. And against the estimate that synthetically accessible drug-like molecules exceed 10^60 [8], every screening claim, this one included, samples a fraction on the order of 10^-49 [3]. That is the honest frame for library size as a selling point.
Which is why the co-crystals do more work than the 5.6-million-CPU run. Near-linear scaling in the AWS cloud is a real engineering result [5], but it is the compute reduction, not the millions of cores, that makes the capability portable. The structures of FSP1 complexes, which the authors say gave mechanistic insight into inhibition [7], are the part that indicates the docking rank corresponded to physical binding. Two targets, FSP1 and PARP1 [6], is a demonstration rather than a hit rate, and "nanomolar" spans three orders of magnitude; the abstract reports no potency values.
The quieter contribution is preparation. The authors ship the Enamine REAL Space in screening-ready form, described as the largest ready-to-dock drug-like library they know of, and additionally in SELFIES [2], which is what turns a vendor catalogue into something a graduate student can queue up. More than 1,500 docking protocols come wired in, including GPU-accelerated and machine-learning methods [4], against a field where many platforms remain CPU-only [14]. Whether that translates into a production win is a different measurement: cost per experimentally confirmed hit, including synthesis of on-demand compounds and assay time. This paper leaves that number unreported, and a compute-reduction figure is no stand-in for it.
Ranked by verification strength, evidence, and original report placement.
AdaptiveFlow is an open-source platform presented as making ultralarge virtual screenings more accessible, scalable and efficient, and as supporting AI and machine-learning method development.
AdaptiveFlow provides a screening-ready version of the Enamine REAL Space, described by the authors as, to their knowledge, the largest library of ready-to-dock drug-like molecules, comprising 69 billion compounds, also available in SELFIES format.
AdaptiveFlow integrates more than 1,500 docking protocols, including GPU-accelerated and machine-learning-based methods.
The platform achieves near-linear scaling on up to 5.6 million CPUs in the Amazon Web Services cloud.
The authors identified nanomolar inhibitors of two disease-relevant targets, ferroptosis suppressor protein 1 (FSP1) and poly(ADP-ribose) polymerase 1.
Co-crystal structures provided mechanistic insights into FSP1 inhibition.
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 31, 2026
Follow any of these and your For You feed starts watching them — no settings page required.
build
The nightly shutdown Lambda earns its postmortem on the morning restart1 distinct publisher
security
ONCD stakes Texas water security on six months of donated vendor red teaming1 distinct publisher
build
Partial batch failures: the handler is fine, the mapping decides1 distinct publisher
product
Workday grew subscription revenue 13.9% on the pricing model agents were meant to replace1 distinct publisher
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Peer-reviewed, but one interested voice
Peer review and a co-crystal structure put a real floor under this. What keeps the score mid-range is that the tool's builders are also its only assessors in our coverage, and the two claims doing the most persuasive work — orders-of-magnitude cheaper screening, near-linear scaling to millions of cores — arrive as summary assertions ahead of the data that would let anyone check them.
Used so far by the people who wrote it
One release, two internal campaigns, no outside user. That is the whole record. Open-sourcing lowers the barrier for others but is not itself uptake, and nothing in our coverage shows a second group docking anything with this platform or downloading the SELFIES library.
Accessibility asserted, scale demonstrated
The gap sits in the word accessible. What was actually shown is a 5.6-million-CPU cloud run; what is promised is screening within reach of a modest lab, and the bridge between them is a compute saving quantified only as orders of magnitude. 'Largest library of ready-to-dock molecules' is also fragile by the paper's own accounting, since 69 billion is a couple of percent of the on-demand space it cites two paragraphs earlier. The overstatement is one of framing, not fabrication — the biology behind it is more solid than the arithmetic around it.
Builders grading their own build
Nobody here is disinterested. The group that wrote the platform chose the targets, ran the comparison, defined 'exhaustive docking' as the thing it beats, and named no rival system it might have lost to. Two commercial dependencies also get flattering mentions without terms attached — Enamine's compound space and Amazon's cloud — and the front matter available to us carries no funding or competing-interest detail. Open-sourcing the code cuts against the obvious commercial motive, which is why this lands short of the high end.
Firm on the chemistry, thin on the engineering
Split the story and confidence splits with it. That inhibitors were found and co-crystallised is well attested and would be costly to get wrong. That the platform makes billion-scale docking affordable for ordinary labs rests on a single unreplicated efficiency claim from an interested party, with no outside user yet to confirm or deflate it.