Skip to content
antibacterial .ai

Use cases

Drug discovery software, and what four teams do with the answer

Four teams, four different reasons to screen a compound before it reaches a plate. The read-out is the same; what changes is the decision it feeds.

Medicinal chemist, 20 to 200 people

Biotech antibacterial program

Four hundred analogs on paper, plate capacity for twelve a month, and a program review in three weeks.

The run

  1. 01 Upload the series as an SDF and run it against ESKAPE.
  2. 02 Sort by how much of the panel each compound holds, then by confidence.
  3. 03 Read the attributed losses: the ones dying to efflux are a chemistry problem, the ones dying to a target mutation are a scaffold problem.
  4. 04 Take the top twelve to the plate and take the attribution to the review.

What comes out

A plate list with a written reason next to every inclusion and every exclusion.

Usual plan
Discovery
Unit of work
Compound times strain
Status
Prediction, research use only

Microbiology or assay lead

Discovery CRO

Client submissions arrive with no triage, so half the panel time goes on compounds that were never going to work on Gram-negatives.

The run

  1. 01 Screen the submission before quoting.
  2. 02 Quote a panel that matches the realistic activity range instead of a default set.
  3. 03 Send the client the predicted matrix alongside the quote so the scope conversation is about data.
  4. 04 Run the plate, then compare the measured result against the prediction for your own calibration.

What comes out

Fewer wasted wells, quotes that survive scrutiny, and a defensible reason to propose a different panel.

Usual plan
Program
Unit of work
Compound times strain
Status
Prediction, research use only

Principal investigator, small budget

Academic AMR lab

A handful of natural product or peptide leads, assay time that has to be applied for, and no informatics staff.

The run

  1. 01 Screen each lead against the panel that matches the grant question.
  2. 02 Use the analog lookup to find what has already been reported nearby, before writing the background section.
  3. 03 Take the two leads with the best predicted Gram-negative coverage into the assay application.
  4. 04 Cite the prediction as triage, never as a result.

What comes out

An application that shows the shortlist was reasoned, not arbitrary.

Usual plan
Lab
Unit of work
Compound times strain
Status
Prediction, research use only

Extract chemistry and dereplication

Natural product screening

Extract libraries are full of known actives rediscovered for the fifth time, and the cost of finding that out is a full assay.

The run

  1. 01 Screen the isolated structure as soon as it is elucidated.
  2. 02 Check the analog list: if the closest published neighbour is a known antibiotic class, that is your dereplication signal.
  3. 03 Prioritise the structures whose neighbourhood is genuinely empty, and accept that those come back at confidence one.
  4. 04 Send only those to the panel.

What comes out

Rediscovery caught at the structure, not at the plate.

Usual plan
Discovery
Unit of work
Compound times strain
Status
Prediction, research use only

For heads of discovery

The budget case, as arithmetic you can check

No invented numbers here. Put your own figures in the left column and the comparison holds or it does not.

See the tiers

One confirmatory MIC panel at a CRO

Hundreds of dollars, about a week of calendar time

The thing a screen is trying to avoid spending twice

Discovery plan

$499 a month, 2,000 screens

Less than one such panel per month

A series carried three months too long

Chemistry time, plate time, and a program slot

The failure a named mechanism is meant to prevent

A wrong prediction

One plate, caught immediately

Cheap, which is why the confidence level matters more than the number

Governance is usually the second question: SSO, roles, an audit log, a DPA, and a written commitment that submitted structures are never used to train anything. The security page has all of it in one table.

Start with the compound your team disagrees about

Run a screen on the homepage without an account. When the read-out is useful, create one and bring the rest of the series.

Run a screen first

What you are agreeing to

  • Your compounds stay yours
  • Never used for model training
  • Deletable on request
  • No card required to run a screen

Screen your own compound.
No card required.