Absci

The molecules are designed by the models. Given a target structure, the de novo model designates a specific epitope and generates antibodies against it, which is a different act from screening a library and keeping what binds: the sequence did not previously exist and was not selected from a set someone else assembled.

The wet laboratory is substantial and it is infrastructure in service of the models rather than a co equal asset. Its purpose is to generate the training data and validate the designs, and the company's own framing puts it in that order: the data to train, the artificial intelligence to create, the laboratory to validate.

That distinguishes this record from the laboratory diagnostics in this index, where the assay and the classifier are genuinely co equal and the grade is held at B.

Low autonomy and the axis sits oddly on a discovery platform, which is worth saying rather than forcing a grade.

The models propose candidate molecules; scientists choose which to synthesise, the laboratory tests them, and every downstream decision runs through the ordinary machinery of preclinical study, regulatory submission, ethics approval and clinical trial protocol. Nothing acts on a patient without that chain.

The meaningful autonomy question in this category is upstream and unexamined: the model chooses which epitope to target and which sequences to advance, so what never gets designed is invisible in a way a rejected screening hit is not.

More mechanism disclosed than most of this lane. The platform's stages are described distinctly rather than as one black box: a de novo foundation model for epitope directed design, separate optimisation models for developability, half life and immunogenicity, and a wet laboratory loop feeding data back.

Throughput claims are specific enough to be checked against a customer's experience, at billions of cells screened weekly and as little as six weeks from design to validated candidate, and the company has published development detail on its lead programme rather than only announcing milestones.

What is absent is the usual: no architecture, no training data description, and no reported hit rate or success rate for the design step itself, which is the number that would let an outsider judge the platform rather than the pipeline.

The pipeline stages are named individually, which is the useful half of this axis for a company of this shape, covering a foundation model for design and separate optimisation models for named properties, so a partner knows which components exist even though none is attributed to a supplier.

The clinical form of this axis fits loosely here and that should be said plainly rather than left to inference: this is a drug developer rather than a handler of health records, and the discovery workflow operates on sequences, structures and laboratory data.

Where patient data does exist is the clinical programme, governed by trial protocol, ethics approval and good clinical practice rather than by the arrangements this axis usually examines, and a reader should not infer weakness from the absence of a data handling position, because it is the wrong instrument for the object. What remains askable is the partner side and the trial side separately.

On the partner side, no hosting arrangement or sub processor list was located, and nothing states whether partner supplied targets, sequences or assay results contribute to models used for others, which is the commercially sensitive question in this lane. On the trial side, ask who holds participant data, for how long, and whether anything derived from it reaches the discovery platform, since a company that both runs trials and trains models has a path between them that nothing describes.

The strongest evidence position available in this category short of an approved drug, because the molecules have reached people.

Most of this lane is evidenced by discovery speed, hit rates or partnership value, none of which says whether the resulting molecule works. Here two artificial intelligence designed antibodies have interim Phase 1 data: one reporting an extended half life relative to first generation competitors in its class along with a favourable safety profile, the other reporting safety, pharmacokinetic and immunogenicity results from a trial of up to 227 participants.

Held at B rather than A because none of it is efficacy. Phase 1 establishes that a molecule behaves as designed in humans and is tolerated, not that it treats the disease, and the proof of concept readouts are still ahead. The honest reading is that this validates the design method, not the drug.

Graded on an honest basis, and the axis fits loosely because this is a drug developer rather than a handler of health records.

Where patient data does exist is the clinical programme, governed by trial protocol, ethics approval and good clinical practice rather than by the arrangements this axis usually examines. No position on data handling was located, and a reader should not infer weakness from that: it is the wrong instrument for the object.

Graded on an honest basis and largely inapplicable. A company running its own trials and designing molecules is not a business associate of a provider organisation; its obligations run through trial regulation and its agreements are with investigators, sites and pharmaceutical partners.

Recorded so the record is complete rather than because a gap has been found.

Recorded honestly and provisionally: the dedicated trust and security search this index requires was not run in this pass, and no attestation was encountered incidentally.

As a listed company it carries mandatory cybersecurity risk management and governance disclosure in its annual filing, which is the authoritative source and was not checked. For a platform company the asset at risk is the model and the proprietary data behind it rather than patient records, which is a different threat model from most of this index.

Read as a drug developer rather than a device maker, which is the correct frame for this category.

The position is real: investigational new drug enabling studies completed for the lead antibody, first participants dosed, and a second programme running a Phase 1 and 2a trial in Australia in up to 227 participants. Clearing an investigational new drug application and dosing humans is a regulatory achievement of a different order from a preclinical platform claim, and much of this lane has not reached it.

Held at B because there is no approved product and no pivotal trial, and because the company has signalled it may out license its lead rather than carry it through registration, which means the regulatory path beyond Phase 1 may be run by somebody else.

The usual framing of this axis does not transfer, and the substitute question is a real one that the company addresses.

For a designed biologic the safety issue created by the artificial intelligence is immunogenicity: a sequence the human immune system has never encountered may provoke a response against itself, and the more novel the design the more that matters. The company states it optimises for reduced immunogenicity risk at the design stage and reports immunogenicity results in its interim Phase 1 data, which is the right disclosure for this category and better than silence.

What is not published is any systematic evaluation of the platform: how often designed candidates fail on immunogenicity, developability or manufacturability, and whether that rate differs from conventionally discovered antibodies. Selection of trial populations and generalisability of results across them is also undescribed.

The platform's stages are described distinctly rather than as one black box, which lets a reader locate where a failure would occur. A foundation model performs epitope directed design, separate optimisation models handle developability, half life and immunogenicity, and a wet laboratory loop feeds data back.

Naming the optimisation objectives individually matters, because those three pull against each other and against binding affinity, so knowing they are separate models tells a partner that trade offs are being made explicitly rather than absorbed into one score.

Throughput claims are also specific enough to be checked against a partner's own experience, at billions of cells screened weekly and as little as six weeks from design to validated candidate, and the company has published development detail on its lead programme rather than only announcing milestones. Held at C because the number that would let an outsider judge the platform rather than the pipeline is absent.

No hit rate or success rate is reported for the design step itself, and the distinction is the whole assessment: pipeline progress tells a reader that some programmes worked, which is subject to exactly the selection effect that makes successes easy to publish, while a design step success rate with a denominator says how often the technology delivers. Ask for the proportion of designs that meet their specification on first synthesis, the denominator behind it, and how that has changed over time.

Not applicable in any meaningful sense and graded neutrally rather than penalised. A discovery platform does not connect to a record system; its data flows are between laboratory instruments, models and trial systems.

Recorded so the axis set stays complete across the index. A reader comparing this record against a care delivery vendor should discount this line entirely.

Not applicable in the usual sense: nothing is deployed into a customer environment, since the company designs molecules rather than shipping software.

What does exist is a physical and computational footprint across three locations, a headquarters and laboratory in Washington state, an artificial intelligence research laboratory in New York and an innovation centre in Switzerland, which is where the work happens rather than where a customer's data rests.

No product is on sale, so the usual questions do not apply, and listing produces disclosure of a different kind: revenue, cash position and programme spend are all public, and analyst coverage exists.

The commercial model is dual and only half of it is visible. Partnered programmes with pharmaceutical and technology companies carry undisclosed per deal economics, and the internal pipeline is pre revenue. The strategically important disclosure is the stated intention to explore partnership or out licensing for the lead antibody rather than developing it further internally, which tells a reader where the company expects value to be realised: at design and early clinical proof, not at commercialisation.

One modality, several indications, and no care setting, which is what this axis measures for a discovery company.

The platform designs biologics, principally antibodies, and the pipeline spans inflammatory bowel disease, androgenetic alopecia, endometriosis, an undisclosed immuno oncology target and HER2 positive disease. That spread demonstrates the platform is not target specific, which is the point being proven.

Worth noting plainly rather than as criticism: the programme furthest along in generating human efficacy data targets pattern hair loss, a large commercially attractive indication with real psychological burden and no mortality. For a platform company proving a design method, choosing a large accessible market with a fast readout is a rational capital decision, and it does mean the technology's first human proof point will come in a consumer adjacent indication rather than a life threatening one.