Preclinical valuation: what sits behind the number

1. What the headline number is#

The range on the results page is a therapeutic-area × phase baseline multiplied by the modifiers you set (modality, territory, competitive position, data quality, designations). The baseline comes from a live calibration when at least five verified, disclosed deals exist for that area and phase, and from the static benchmark tables otherwise; the results page says which. The low–high band is a fixed directional width around the median, not a percentile of the comparable set.

Below Phase 2 the rNPV family has little signal: cumulative success probabilities of a few percent and a decade to market make the risk-adjusted NPV a small residual of two large numbers. So in the ensemble the comparable-transaction method is floored at two-thirds of the blend for discovery, preclinical and Phase 1 assets. In plain terms: at this stage the number is anchored on what buyers paid for similar programs, and the rest of the model widens or narrows the band around that anchor.

2. Which comparable deals, and how many#

Every deal in the database is scored against your inputs with the additive rule below (the same constants the product reads). A deal qualifies when it shares the therapeutic area and at least one of same phase, adjacent phase or indication. If fewer than 5 deals qualify, the rule widens in two steps, to area + modality and then to area only, and the panel labels the widened pool. The top 30 by score are kept; upfront and total ranges are recency-weighted quartiles with a 2.5-year half-life.

Only rows with disclosed terms are eligible, and rows the verifier has rejected or flagged are excluded. Rows still awaiting verification are included and labelled. On a Pro seat the Deal Transparency panel lists up to fifty of these comps with upfront, total value, phase, modality, date, verification status and a link to the filing or release, and exports them to CSV. That list is the answer to “which deals is this built on”.

The honest part is the size of the early-stage pool. After the September 2026 verification sweep, the preclinical and discovery deals that pass the filter above number 617 across all areas (live, 2026-10-08):

Why the pool is small: preclinical comparables are held to the strictest standard in the database. A deal must have disclosed terms, be confirmed against a primary source, and carry no unresolved discrepancy on the money; a confirmed deal whose announced date or milestone total still needs correction waits outside the pool until the corrected figures and citation are applied. That standard is deliberate, because at this stage the comparable list is the artifact a committee will examine. The pool is growing on two fronts from late September 2026: confirmed deals are being brought in with corrected figures and citations, and sourcing now targets preclinical transactions with disclosed upfronts specifically. This table updates every fifteen minutes; the number you see is the number the product uses.

3. How peak sales is estimated before clinical data#

Peak sales is estimated at the level of the indication, not the asset, and by the same method at every phase. What changes at preclinical is how much it matters (little, see section 1) and how heavily it is risk-adjusted (section 4). The order of precedence in the engine:

1. Your number. Enter an analyst or internal peak-sales view and it drives both the headline (as a separate, labelled adjustment) and the rNPV.
2. Epidemiology-derived estimate. For about 270 indications: prevalence × diagnosed × treated × drug-eligible × adoption ceiling × share by competitive position × annual cost × territory price index, cross-checked against a curated range from company filings and industry forecasts, and hard-capped at 80% of the indication’s addressable market.
3. Multiple of deal value. When neither exists: for preclinical, 6× / 12× / 20× the benchmark deal value (low / median / high).

The default shown in the calculator’s peak-sales field is this same estimate, read from the same epidemiology path and TAM cap the rNPV runs on, so the number you see is the number the model uses. Your own figure overrides it in both places. (Until Sep 25 2026 the field showed a separately curated “typical asset peak”, which could differ from the engine’s figure.)

4. The preclinical risk chain#

A preclinical asset carries the full chain of transition probabilities to approval (base rates by therapeutic area from BIO, Citeline and Wong–Siah–Lo, adjusted for modality, biomarker strategy and designations), the full development timeline, and a stage-specific discount rate. Read live from the engine for a small molecule with no biomarker selection and no designation:

So a preclinical program is valued at roughly 3%–12% of its unrisked value before discounting. On top of that, the Monte Carlo layer uses wider early-stage scenarios (bear peak sales −40% and bull +50%, with heavier tail weights than at later stages), and the implied upfront is taken as a small share of rNPV at this stage (3–8%, engine 5.1.0). Those two constants live inside the engine rather than in a shared table, which is why they are quoted with a version.

5. What the backtest shows for preclinical, and what it does not#

We publish a backtest of predicted against disclosed upfronts on /accuracy. Three things to know before reading its preclinical slice. It scores the rNPV engine, not the calculator’s headline baseline. The test harness applies an early-stage floor to predictions that the product does not apply. And the held-out split is by deal id, not by time, and was introduced after early rounds were tuned on the full corpus, so it is a partial check rather than a clean out-of-sample test.

Read plainly: at preclinical the engine’s upfront lands within a factor of two about as often as the phase median alone does, and many of its predictions sit at the harness floor. That is the empirical reason the product anchors preclinical on comparables and shows a range rather than a point: no cash-flow model carries much information about a specific program’s upfront at this stage, while the comparable set does. Total deal value is a different question at preclinical, because milestone stacks are set by program count and strategic intent, which is why the panel shows the comps’ milestone structures rather than predicting one. Separately, a per-factor regression on the same corpus finds preclinical deals close at roughly 2.9× what the Phase 2 baseline would imply (95% CI 2.2–3.8×), and the early-stage baselines reflect that.

What this means for use: treat a preclinical output as a range plus a comparable list, not a point estimate. The comparable list, with sources, is the defensible artifact; the band tells you where in that list a program with your profile tends to sit.

6. What is changing#

• Confirmed early-stage deals awaiting a corrected upfront, total, date, stage or asset name are being brought into the pool with the corrected figures and a citation, early-stage rows first (from Sep 25 2026).
• Ingestion now asks specifically for preclinical and discovery deals with disclosed upfronts, 2019 onward, across every therapeutic area, and SEC full-text search includes research-collaboration, discovery-collaboration and option-to-license agreements.
• The calibration tables refresh weekly; when an area reaches five verified preclinical deals its baseline switches from the static table to the live one, and the results page says so.
• The deal-terms backtest now runs on the first of every month and lands as a reviewed update; a time-based holdout is the next methodology item.

Questions about any figure on this page go through the contact page. Every number above is read from the engine, its data files or the database at render time.