AI for Bio · Model

Beacon (Inductive Bio)

Inductive Bio's closed ADMET/PK predictor trained on a pre-competitive pharma data consortium

Small moleculesPlatform companyUnited StatesProduct only, no public artifact中文

Overview

NameBeacon (Inductive Bio) · aliases: Beacon-2, Inductive Bio Beacon
Full nameA product name with no expanded form
Ambiguous names (not auto-linked)Beacon
FamilyBeacon is the ADMET model; Beacon-2 adds human dose estimation, and the challenge wins belong to Beacon, not Beacon-2
  • Beacon · †
  • Beacon-2 · †
OrganizationInductive Bio
Organization typePlatform company
CountryUnited States
Model classProperty prediction (ADMET / PK)
DomainSmall molecules
First release† partly verified · product update
Basis: Press coverage of the company with its Beacon platform (model release date not given; month approximate) source
Other milestones
  • First public evidence: † · Press coverage source
  • Official announcement: † · Beacon described in Series A coverage source
Latest update† partly verified · successor
Basis: Beacon-2 launch (secondary report)
Beacon-2 adds human efficacious-dose prediction; a medicinal-chemistry agent, Indy, was introduced days earlier
Note: The original press release was not opened; date from secondary reports source
The most recent sourced event, not a claim that nothing changed since.
StatusProduct only, no public artifact · Closed commercial product with successor Beacon-2
OpennessCode: Closed
Weights: Closed
Closed; provided to partners through the company platform
Benchmarks reported0 counted (unconfirmed 1)
Wet-lab validationNone recorded
Unverified fields† first_release, latest_update, release_milestones.first_public_evidence, training_data.c_own_wetlab_for_training, benchmarks_reported[0]
As of2026-11-17

What it does

Predicts ADMET and pharmacokinetic properties before a compound is made, fine-tuned for each partner's chemical space; Beacon-2 adds human dose estimates. It shares the ADMET route with ADMET-AI and MiniMol, but its training data come from a consortium pooling IP-protected data from several companies.

Benchmarks reported

Every row is one pairing from the survey's relation file; a score belongs to the benchmark page's ranking, where it is shown with metric, eval set, source type and date. 0 of 1 pairings are counted.

BenchmarkRelationEval set / protocolSource typeNote and evidence
27 Polaris / ASAP / OpenADMETUnconfirmed (not counted)—Unconfirmed sourcePress release only; organiser anonymises entries; model identity unverified †disputed · evidence

Other evaluations (outside the 44)

  • OpenADMET blind challenges (Public benchmark not among the 44) Self-reported: The press release claims first place in antiviral, ExpansionRx and PXR challenges; only one is echoed by a third-party write-up and no official final board was read
  • Beacon-2 dose estimation (Own task set) Self-reported: Dose prediction on ExpansionRx public compounds (secondary reports); sources give inconsistent figures
  • Partner fine-tuning (user or partner report) Self-reported: Fine-tuned per partner on consortium data; scale not disclosed

Developers report: repeated first places in blind ADMET challenges, with consortium data and partner fine-tuning capturing subtle structural signals; Beacon-2 adds human dose estimates.

Independent evidence: No independent evidence found. The challenges are organiser-run, but only press releases and secondary reports were read; the boards are pseudonymous and no matching entry was verified

Training data

Group in Figure 7: Private or licensed data.

ClassValueDetail
a. Public databasesYesPublic data used; libraries not named source
b. Private or licensed dataYesA pre-competitive consortium pooling anonymised partner data; compound and assay counts unpublished source
c. Own wet-lab data used for trainingNot disclosedNo statement of own experimental data
d. Wet lab used only for validationNoBlind challenges are external tests, not own wet lab

Wet-lab validation

None. The developers report no wet-lab validation.

Background

  • Organization: Inductive Bio is based in New York and does not develop its own drugs; it commercialises a pre-competitive consortium where companies contribute anonymised data under IP protection and clients pay software and collaboration fees. It left stealth in 2023-12.
  • Funding: About $4.3M seed disclosed at launch (a16z Bio + Health and Lux Capital); a $25M Series A led by Obvious Ventures announced in 2025-05.
  • Products and partnerships: Products are the Compass platform, the Beacon models and the chemistry assistant Indy; Denali Therapeutics is an early partner; with Ginkgo Datapoints and Tangible Scientific it forms an assay, modelling and compound-management workflow.
  • Related models: ADMET-AI, MiniMol

Reading notes

  1. First places suggest strong results on those datasets, but the evidence is company press releases and the boards are pseudonymous.
  2. Consortium data size is undisclosed, so it is unclear whether the edge comes from the model or the data.
  3. "Consecutive wins" is a point-in-time claim and needs a date when cited.
  4. The two second-hand dose-estimation figures for Beacon-2 must not be mixed.

Caveats

  1. The Polaris pairing is not established: identity cannot be verified and two challenges were outside the Polaris hub.
  2. Beacon-2's launch date and description come from secondary reports.
  3. The first-release month is approximate; no public release date exists for the model itself.

Sources

  1. Company press release (PXR challenge) (press)
  2. Beacon-2 coverage (second-hand) second-hand
  3. Series A coverage (press)
  4. Company interview coverage (press)
  5. OpenADMET challenge lessons write-up (third-party paper)
  6. Stealth exit coverage (press)

Corrections

1 correction
FieldWasNowBasisFound
benchmarks_reportedFirst in three Polaris blind challenges (self-reported)Unconfirmed and not counted: the board is anonymous, and two of the events are not on the Polaris platformsource2026-11-16

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-17.