Agentic AI for drug-target ideation

AI agents that build the case for a drug target

A team of AI agents researches your target and writes a report where every claim is sourced and checked.Name a gene or protein and a disease. A lead agent plans the work, ten specialist agents do the research, and every claim in the report links to its source and shows whether it checks out.

Join the waitlistHow we protect your work

The product pictures on this page are illustrative: real gene symbols, made-up results.

 

contradicted · devil's advocate · flagged on the version

Illustrative

00 / Why

Why the right target matters

the target is chosen first, and proven in patients last

Most drug programs fail. About 90% of drug candidates that enter clinical trials never reach approval.[1]

About 9 in 10 drug candidates that enter clinical trials fail, and lack of efficacy is the top reason. Illustrative: the counts at each stage are schematic; the cited figures are in the text.

  • Lack of efficacy, found late

    The most common reason is that the drug doesn't work in patients: lack of clinical efficacy accounts for 40–50% of failures.[1] Sometimes the biology behind the target was wrong, and that is decided years earlier, when the target is chosen.

  • Slow and expensive

    The failure shows up late, often years in. A widely cited estimate puts the cost of an approved drug at $2.6 billion, including the cost of failures (in 2013 dollars, counting the cost of capital).[2]

  • Better evidence up front

    Drug targets with human genetic evidence for the disease being treated are about 2.6 times as likely to succeed from clinical development to approval.[3]

So the cheapest place to stop a bad program is before it starts.

That is where TargetID.ai's agents work. Specialist agents collect the evidence for and against a target, and a devil's advocate agent looks for the claim that doesn't hold. See how it works

Sources · industry statistics, cited

  1. [1]Sun D, et al. “Why 90% of clinical drug development fails and how to improve it?” Acta Pharm Sin B 2022;12(7):3049–3062. doi:10.1016/j.apsb.2022.02.002 (opens in a new tab)About 90% of clinical drug candidates fail; lack of clinical efficacy accounts for 40–50% of failures.
  2. [2]DiMasi JA, Grabowski HG, Hansen RW. “Innovation in the pharmaceutical industry: new estimates of R&D costs.” J Health Econ 2016;47:20–33. doi:10.1016/j.jhealeco.2016.01.012 (opens in a new tab)$2.6 billion per approved drug, including the cost of failures (in 2013 dollars, counting the cost of capital).
  3. [3]Minikel EV, et al. “Refining the impact of genetic evidence on clinical success.” Nature 2024;629:624–629. doi:10.1038/s41586-024-07316-0 (opens in a new tab)Genetic support for a target–indication pair is associated with a 2.6-fold higher probability of success from clinical development to approval.

01 / Ask

From a question to an approved report

Ask · Plan · Research · Write · Review · Approve

Ask about a target and a disease. The agents plan the work, and you can approve the cost first. Specialist agents research in parallel, the report is written section by section, a reviewer agent double-checks every source, and your colleagues sign off.

02 / Investigate

A team of specialist agents, led by one

One lead agent · nine specialists · a reviewer and a skeptic at the end

A lead agent decides which specialists to bring in, and they can bring in help of their own: the literature agent hands papers to a reader. You can watch each one work, and stop everything at any time.

Details +
  • Genetics, biology, clinical data, safety, chemistry, and the competitive landscape, each with its own research tools.
  • They search the major biomedical databases, the published literature (PubMed, Europe PMC and more, read in full when needed) and the web.
  • Before the report is final, a devil's advocate looks for evidence against the target and a reviewer double-checks every source.
  • Genetics, biology, clinical data, safety, chemistry, and the competitive landscape, each with its own research tools.
  • They search the major biomedical databases, the published literature (PubMed, Europe PMC and more, read in full when needed) and the web.
  • Before the report is final, a devil's advocate looks for evidence against the target and a reviewer double-checks every source.

03 / Cite

Every claim is cited, and every citation is checked

Every citation points to real evidence

  1. 01

    Written from evidence

    Agents can't make up a citation. Each one is attached to evidence the agents actually found, and a paragraph that makes a claim without evidence is sent back. A very short sentence can occasionally slip through without a source.

  2. 02

    Checked, source by source

    Every source shows whether it has been checked and whether it holds up. The reviewer goes back over them, and anything the evidence contradicts is marked in red.

  3. 03

    Reasoning kept apart

    When an agent reasons beyond the evidence, it says so, shows what it is based on, and how confident it is. That reasoning never counts toward a score, and missing sources or an empty Risks section are flagged.

04 / Trace

See the evidence behind every report

Grouped by topic · linked to sources · reasoning shown separately

A map of the report's evidence: what supports what, and where each piece came from.

Details +
  • Hover over a point to see the claim and whether it checks out; click to read the evidence.
  • A list under the map stays in sync with it, so it works with a keyboard and a screen reader too.
  • Every saved version keeps its own map, so older versions always look the way they did.
  • Hover over a point to see the claim and whether it checks out; click to read the evidence.
  • A list under the map stays in sync with it, so it works with a keyboard and a screen reader too.
  • Every saved version keeps its own map, so older versions always look the way they did.

05 / Rank

Scores and rankings, to compare targets side by side

Seven scores · your priorities · no hype from thin evidence

Change your priorities and the same evidence ranks differently. The targets here are ranked the way the product ranks them, using made-up scores: watch STAT3 get held back for being strong on only one front.

Details +
  • Each target is scored on seven dimensions, from genetic support to competition. A score backed by little evidence stays close to neutral.
  • The overall score comes with a range showing how sure it is. A target that looks strong on only one front, or has a safety concern, can't rank at the top.
  • Start from Balanced, Genetics first or Fast to clinic, or set your own priorities. The ranking comes from the evidence, never from an agent's opinion.
  • Each target is scored on seven dimensions, from genetic support to competition. A score backed by little evidence stays close to neutral.
  • The overall score comes with a range showing how sure it is. A target that looks strong on only one front, or has a safety concern, can't rank at the top.
  • Start from Balanced, Genetics first or Fast to clinic, or set your own priorities. The ranking comes from the evidence, never from an agent's opinion.

06 / Control

You direct the agents: what they do, what they spend, and what they look at next

Plan first, then approve the cost

See the plan and the price before anything runs

Turn on Plan mode and the lead agent proposes a plan instead of starting work. Nothing runs until you approve it.

Details +
  • Each step is a specialist agent with a goal, and steps that can run at the same time do.
  • The estimate is a dollar range based on your team's past work of a similar size.
  • Only the agents in the plan can run, plus the reader, the reviewer and the devil's advocate, which are always on hand.
  • Each step is a specialist agent with a goal, and steps that can run at the same time do.
  • The estimate is a dollar range based on your team's past work of a similar size.
  • Only the agents in the plan can run, plus the reader, the reviewer and the devil's advocate, which are always on hand.

Each agent role, on the provider you choose

Anthropic · OpenAI · Google · your own

Pick the AI provider for each kind of agent. Your keys are encrypted and never shown again in full.

Details +
  • Use Anthropic, OpenAI, Google or your own compatible service, all at once or one per role.
  • Each choice is checked to make sure it can do the job, so nothing starts half set up.
  • Separate workspaces for each team, tags to organize targets, and two-factor sign-in.
  • Use Anthropic, OpenAI, Google or your own compatible service, all at once or one per role.
  • Each choice is checked to make sure it can do the job, so nothing starts half set up.
  • Separate workspaces for each team, tags to organize targets, and two-factor sign-in.

Select a sentence, and ask about it

Ask, verify, expand or challenge any sentence

Put the agents back to work right on the report, next to the claim, not in a separate chat.

Details +
  • Ask, Verify, Expand or Challenge sends the sentence to the agents; a Note is just for your team.
  • A discussion stays attached to its sentence, and tells you if that sentence changes.
  • Agents can't rewrite the report on their own. They suggest edits, and you accept or reject them.
  • Ask, Verify, Expand or Challenge sends the sentence to the agents; a Note is just for your team.
  • A discussion stays attached to its sentence, and tells you if that sentence changes.
  • Agents can't rewrite the report on their own. They suggest edits, and you accept or reject them.

Hover over a compound, see the molecule

Structure · formula · stage of development

Point at a ChEMBL compound ID anywhere in a reply or a report, and its structure and key facts appear right next to the text.

Details +
  • The details come from ChEMBL, the public drug database the agents use.
  • If an ID isn't a compound, it says so instead of guessing.
  • Works on a phone too: just tap the ID.
  • The details come from ChEMBL, the public drug database the agents use.
  • If an ID isn't a compound, it says so instead of guessing.
  • Works on a phone too: just tap the ID.

07 / Scale

Many targets, one plan, one budget

Up to 100 targets · one estimate · a spending limit you set

Scroll to run the batch below: it works through three targets at a time, just like the real thing.

Details +
  • Paste a list or upload a spreadsheet of up to 100 targets, each with its own team of agents.
  • You see one estimate for the whole batch, and nothing runs until you approve it. It stops starting new targets if spending runs well past it.
  • Pause, resume or retry at any time, then get a side-by-side comparison and a combined report.
  • Paste a list or upload a spreadsheet of up to 100 targets, each with its own team of agents.
  • You see one estimate for the whole batch, and nothing runs until you approve it. It stops starting new targets if spending runs well past it.
  • Pause, resume or retry at any time, then get a side-by-side comparison and a combined report.

08 / Approve

Review, approve, export

Draft · In review · Approved

Every report moves from draft to review to approved, and the board shows where each target stands.

Details +
  • Send a version to reviewers on your team. It can only be approved once the checklist is complete and every comment is resolved.
  • Approved reports are stamped with the reviewer's name and export to PDF, PowerPoint, web page or text.
  • See all your targets as a table or a board, grouped by status, project or tag, and export them to Excel.
  • Send a version to reviewers on your team. It can only be approved once the checklist is complete and every comment is resolved.
  • Approved reports are stamped with the reviewer's name and export to PDF, PowerPoint, web page or text.
  • See all your targets as a table or a board, grouped by status, project or tag, and export them to Excel.

09 / Protect

Your papers stay where you put them

Only Anthropic's models read your files

Add papers and data files to any project. To be upfront about the limits: what an agent writes in its own words after reading your file isn't tracked, and files aren't virus-scanned.

See every security measure

Details +
  • Your files are only ever read by Anthropic's models. Other AI providers never see them.
  • Once an agent has read one of your files, it can no longer search the web, so nothing from it can leak out in a search.
  • Every PDF is checked and cleaned before it is stored, and anything suspicious is rejected.
  • Your files are only ever read by Anthropic's models. Other AI providers never see them.
  • Once an agent has read one of your files, it can no longer search the web, so nothing from it can leak out in a search.
  • Every PDF is checked and cleaned before it is stored, and anything suspicious is rejected.

10 / Begin

Put the agents on your next target

We are opening access gradually. Join the waitlist and we'll tell you when your workspace is ready, and as the agents learn new tools and skills.

Join the waitlistRead the security page