Terray Therapeutics

Phase A — Understand the business

Lens 1 · Company Overview

Terray Therapeutics is a chemistry-first, AI-native small-molecule drug-discovery company founded in 2018 by brothers Jacob Berlin (CEO) and Eli Berlin (Chief Business & Financial Officer), spun out of City of Hope, where Jacob ran an NIH-funded nanotechnology lab. HQ is Pasadena, California, with a lab facility in Monrovia; headcount is ~58.

The business has two engines off one platform:

1. Internal pipeline — a portfolio of wholly-owned, immunology-focused small-molecule programs, all built from chemistry starting points "discovered de novo" by the platform.
2. Partnered discovery — Terray runs its platform against targets nominated by pharma partners, who take molecules forward in exchange for upfront + milestone + royalty economics. Named partners: Bristol Myers Squibb (Dec 2023), Calico Life Sciences (Alphabet's longevity subsidiary), Gilead (Dec 2024), and Odyssey Therapeutics (transcription-factor / IRF5 collaboration).

The platform itself is branded tNova (the integrated AI+experimentation stack at the time of the Series A/B) and, as of Nov 2025, EMMI — "Experimentation Meets Machine Intelligence" — which couples its proprietary ultra-dense microarray hardware and an automated lab with a full-stack AI layer (Generate / Predict / Select models).

Contract structure / payment terms. Partnered deals are the classic platform-biotech shape: an upfront payment + preclinical/clinical/sales milestones + tiered royalties on net sales; partner controls development/commercialization on opt-in. Financial terms are undisclosed in every deal — the only quantified data point is that the Gilead deal was expected to cut Gilead's 2024 EPS by ~$0.01, implying a modest upfront (single- to low-double-digit $M, `` — a $0.01 Gilead EPS hit ≈ ~$13M pre-tax on ~1.25B shares, but that line item bundles multiple deals so treat as a loose ceiling).

Lens 2 · Supply Chain (→ research, data & CDMO inputs)

A drug-discovery platform's "supply chain" is its inputs to data generation and its path to a dosed molecule. Named stakeholders along Terray's chain:

• Upstream — compute & models: NVIDIA is the keystone input. Terray uses NVIDIA DGX Cloud to train its chemistry foundation models, and some models are slated for the NVIDIA BioNeMo cloud service. NVIDIA's VC arm NVentures is also an equity holder (Series B co-lead) — so NVIDIA is simultaneously supplier, investor, and distribution channel. This is a genuine single-vendor compute dependency (chokepoint #1).
• Upstream — proprietary data hardware: Terray's ultra-dense microarray — a nickel-sized chip with 32 million wells, each holding a molecule of interest — is the single-source data generator. It produces >1 billion new target-molecule measurements per quarter. This is built in-house; it is the company's irreplaceable proprietary input (chokepoint #2 — but a self-owned one, which is the good kind).
• Midstream — Terray itself: automated wet lab in Monrovia turns array reads into the binding dataset that trains COATI/TerraBind.
• Downstream — pre-clinical → clinic: Terray has no in-house manufacturing or clinical infrastructure disclosed; like all small biotechs it will depend on CDMOs for GMP supply and CROs for trials. None are named publicly — n/a — private, not disclosed.
• End customers (two kinds): (a) pharma partners (BMS, Gilead, Calico, Odyssey) who consume discovery output; (b) eventually patients / payers for any internal asset that reaches market — years away.

Chokepoint read: the dangerous single-source dependency is NVIDIA compute, not the science. The microarray is owned; the foundation models are theirs; but training at frontier scale rides on one cloud relationship that is also a shareholder.

Lens 3 · Competitive Advantages (moats)

Terray's moat thesis is proprietary data scale, and it is the most defensible-sounding claim in the file:

• The data moat (real, and quantified). Terray has measured 13 billion target-molecule interactions to date (up from 5B at the 2024 Series B, ~2B at COATI's 2024 publication), claimed at ~50× all publicly available chemistry data and doubling annually. In a field where the consensus critique is "garbage in, garbage out" because public bio/chem data is sparse and noisy, owning the largest clean, internally-generated, standardized binding dataset is the single most credible moat a small-molecule AI company can have. This is the bull case in one sentence.
• Foundation-model IP. COATI (peer-reviewed, J. Chem. Inf. Model. 2024, 64(4):1145-1157) is a multimodal contrastive encoder-decoder of chemical space — now 3rd generation, trained on >1B molecules. TerraBind predicts potency without solving binding pose, claiming ~20% accuracy over Boltz-2 and 26× more chemical space screened per unit compute. Caveat: these performance numbers are self-reported, arXiv-preprint, not peer-reviewed, no independent benchmark — strong if true, unverified for now (see Lens 10/13).
• Switching costs / network effects: weak. Pharma partners can and do run multiple AI vendors in parallel; there is no lock-in. The moat is the dataset, not the relationships.
• Bargaining power: low vs. pharma (partners hold the capital, the targets, and the commercialization; Terray is one of 530+ AI-discovery shops competing for those deals ). High vs. its own data — nobody else can cheaply replicate 13B proprietary measurements. The asymmetry: Terray's edge is upstream (data), but the value capture is downstream (pharma), where it is the weaker party.

Moat verdict: the data-generation flywheel is a genuine, rare, quantifiable advantage. Whether it translates into clinical winners — the only thing that monetizes it — is unproven. A bigger, cleaner dataset reduces discovery risk; it does not touch the ~90% of attrition that happens in human trials.

Lens 4 · Segments (→ revenue mix N/A; program & partner mix)

No segment financials exist — segments.csv is empty; Terray is private and pre-revenue. The meaningful "segmentation" is internal vs. partnered programs and by therapeutic area:

• Internal: wholly-owned immunology portfolio (specific targets undisclosed). This is where equity value concentrates if the platform works.
• Partnered: BMS (multi-target, undisclosed areas), Gilead (multi-target, partner-selected), Calico (age-related disease), Odyssey (transcription factors / IRF5, inflammatory disease). These are non-dilutive funding + validation, but the upside is capped (milestones + royalties, not full ownership).
• Geography: single-site US (CA). No geographic segmentation.

All ; no numbers exist for this name. Trend: the mix is shifting toward partnered breadth (four partners signed 2023-2024) while internal programs march toward a first IND — the standard "fund the platform with deals, build equity with internal assets" playbook.

---

Phase B — Measure performance (+clinical / +private overlays applied)

Lens 5 → Pipeline by phase + Funding & valuation trajectory

There is no earnings print to analyze. Under the composed overlay, Lens 5 becomes the asset table and the funding ladder.

Pipeline (by phase) — ``, undisclosed-target caveat throughout:

Disambiguation (important). Public search frequently conflates Terray's lead with the RIPK2 inhibitor OD-001 for IBD — that asset belongs to Odyssey Therapeutics (Phase 2a UC data, two trials slated H2-2026), not Terray. The Terray-Odyssey collaboration is on transcription-factor biology (IRF5), not RIPK2. Terray's own lead target is not publicly named — treat any specific Terray clinical target claim as unverified. n/a — not disclosed.

The single most important pipeline fact: Terray has zero molecules in human trials as of mid-2026, and its own guided clinic date slipped a year (2026 → 2027). That slippage is the central bear data point.

Funding & valuation trajectory — ``:

Series A syndicate: Madrona, Two Sigma Ventures, Digitalis, KdT, Goldcrest, XTX Ventures, Sahsen, Greentrail, Alexandria. Series B added Bedford Ridge (lead) and Maverick Capital alongside the inside round. The $600M post on a pre-clinical platform is the valuation the market must grow into.

Lens 6 → Founder/management signal (sentiment proxy — no earnings calls)

No earnings calls exist (private). The sentiment proxy is founder messaging + partner cadence:

• Berlin's consistent thesis (stable across 2022→2025): "Knowledge of what causes disease has exploded in the 'omics' era, but the ability to discover molecules to treat it hasn't kept pace" — i.e. the bottleneck is chemistry data, and Terray owns the fix. The framing has not drifted — a credibility positive.
• Tone shift to watch: the messaging pivoted from "to the clinic by 2026" (Series B, Oct 2024) to a softer 2027 clinic timeline by 2026, and the Nov 2025 EMMI launch re-emphasized platform & partnered breadth over a named internal clinical candidate. Reading the tea leaves: when a pre-clinical platform re-brands the platform (tNova → EMMI) and leads with partner logos rather than an IND date, it can signal the internal pipeline is taking longer than hoped — or simply that platform marketing precedes a capital event. Either way, the absence of a named, dated internal IND is the conspicuous silence.
• Partner cadence is genuinely strong: four pharma deals (BMS, Calico, Gilead, Odyssey) in 24 months, plus a disclosed BMS milestone achievement — that is third-party validation money, the best kind for a private.

Lens 7 → Catalyst calendar + mechanism/peer comps (+ cap-table & secondary marks)

Catalyst calendar (what de-risks or kills the equity):

• 2027 — first internal IND / clinic entry (the make-or-break catalyst; already slipped once).
• Ongoing — partner milestones (BMS already hit one; each disclosed milestone is a validation tick and non-dilutive cash).
• Next 12-18mo — Series C or a crossover round (see cap-table read); a tier-1 crossover entry would be the IPO-proximity tell.
• Any peer-reviewed TerraBind benchmark — would convert the self-reported +20%/26× claims into a real moat datapoint.

Cap-table & secondary marks (+private lens):

• Syndicate quality: high. Madrona (lead A), Two Sigma Ventures, Digitalis, Alexandria (the biotech-real-estate strategic), Maverick Capital, and — the standout — NVIDIA/NVentures as a strategic. Bedford Ridge (B lead) is a newer, lower-profile name.
• Crossover-fund tell: NOT yet present. No Fidelity / T. Rowe / Coatee / Wellington entry is disclosed. Per the +private playbook, the absence of a crossover fund means the IPO is not imminent — a $600M post-money pre-clinical platform is too early for the public-market crossover money that front-runs an S-1.
• Secondary marks: Terray shares trade on Forge, Nasdaq Private Market, EquityZen, and Hiive — i.e. there is accredited secondary demand, but no disclosed markup/markdown to read. n/a — not disclosed.

Mechanism / valuation peer comps — ``, multiples labeled or n/a:

P/E, EV/Sales, ROE, dividend yield: n/a / not applicable — Terray is private and pre-revenue; peer multiples for the public names were not pulled to a sourced figure and are deliberately not fabricated. The honest comp is the private-round valuation ladder, where Terray's $600M sits an order of magnitude below Isomorphic's $2.1B-round company and below a now-public Insilico — appropriate for a company with zero clinical assets versus peers that at least have a molecule in (or near) humans.

Lens 8 → Stock-Price / value catalysts (private → funding & validation events)

No public stock, so the "value-moving events" are private-valuation and validation catalysts. The pattern over 2022-2026 shows the market rewards data-scale milestones and partner deals, not clinical progress (because there is none yet):

• Up-markers: Series A $60M (Feb 2022, platform launch); NVIDIA equity (Nov 2023, compute + halo); BMS deal (Dec 2023, first big-pharma validation); Series B $120M @ $600M (Oct 2024, NVIDIA doubles down); Gilead deal (Dec 2024); EMMI launch + 13B measurements + BMS milestone (Nov 2025).
• The conspicuous non-event: no clinical catalyst, and a slipped clinic date — the only category of catalyst that re-rates a biotech from "platform multiple" to "drug multiple" has not arrived. Public peers show exactly this: Insilico re-rated on Phase 2a PoC; Schrödinger's potential re-rate hinges on Phase 1 internal data. Terray has not yet earned that gear.

---

Phase C — Judge people & books

Lens 9 · Management

• Jacob Berlin (CEO, co-founder). PhD organometallic chemistry, Caltech under Nobel laureate Bob Grubbs; postdocs MIT + Rice; BA chemistry Harvard (magna cum laude). Ran an NIH-funded lab as associate professor at City of Hope; 40+ publications, 20 patents, 11k+ citations. Archetype: scientist-founder, and an unusually credentialed one — the microarray is his academic work commercialized. The risk of the archetype: deep-tech founders can over-index on platform elegance vs. clinical execution.
• Eli Berlin (co-founder, Chief Business & Financial Officer). Runs business + finance; the commercial/dealmaking counterweight to Jacob's science. Brother-founder pairing — tight alignment, but concentrated control.
• Bench: Narbe Mardirossian (CTO), Vanessa Taylor (CSO), Feroze Ujjainwalla (Chief Business Officer), Anna Goranson (Chief People Officer). The CBO hire signals a deliberate partnering build-out.
• Board / governance (a genuine strength): Bassil Dahiyat (founder/CEO of Xencor — a rare repeat biotech builder), Wendy Young, PhD (30+ yrs drug discovery, ex-Genentech small-molecule leadership; added Sept 2025), Sid Shenai (Bedford Ridge, B lead), plus Chris Picardo, Dusan Perovic (Two Sigma), Sudha Parasuraman, Samuel Bjork. SAB: Elliott Levy, MD (ex-Amgen head of development) added Apr 2025. This is a board that knows how to take small molecules through development — it materially de-risks the "scientists who can't run trials" worry.
• Capital allocation: raised >$200M, deployed into platform + IND-enabling work + partner-funded discovery. Non-dilutive partner cash (4 deals) offsets burn — disciplined for the stage. No buybacks/dividends (N/A). Insider ownership: founder-led, VC-backed; specific % n/a — private, not disclosed.
• Red flags (management): none material surfaced. Brother-co-founder concentration is a minor governance note, well-mitigated by an independent, heavyweight board.

Lens 10 · Forensic Red Flags (re-pointed: data integrity, going-concern, claim verification)

No audited financials exist — standard forensic income/balance/cash-flow analysis is n/a — private, unaudited. The forensic lens re-points to claim integrity and runway:

• Unverified performance claims (the real flag). The headline platform numbers — TerraBind +20% over Boltz-2, 26× compute efficiency — are self-reported, in an arXiv preprint that is "not yet peer reviewed," with no independent benchmark and no disclosed validation dataset. The 13-billion-measurements and 50× public data figures are company-stated and unaudited. COATI itself is peer-reviewed (JCIM 2024) — a real credibility anchor — but the newer, more impressive efficiency claims are not. Treat the marketing metrics as directionally credible, quantitatively unverified.
• The "de novo discovery" claim. AI-discovery literature is explicit that the proof is in clinical translation, not benchmark accuracy — DSP-1181 (Exscientia/Sumitomo) was AI-discovered, sailed through preclinical, and failed in Phase 1. Terray's de-novo claims are unfalsifiable until a molecule is dosed.
• Going-concern / runway. Not disclosed. $120M Series B (Oct 2024) + partner milestones against a ~58-person burn; `` runway into ~2026-2027 absent a raise, which lines up with the need for a Series C around the 2027 clinic entry. No going-concern signal either way — n/a — private.

Regulatory findings (required sub-section). Per companies/terray/regulatory/regulatory-findings.md (generated 2026-06-30, sources SEC EDGAR EFTS LR + AAER): total_sec_findings: 0. Terray has no CIK and is not an SEC registrant — no EDGAR enforcement search is possible. Non-SEC web search ("Terray Therapeutics" (FTC OR DOJ OR FDA OR... ) enforcement) returned no material enforcement, litigation, consent decree, fine, or penalty. No 10-K Item 3 exists (private). Conclusion: no material regulatory or legal findings — verified via SEC EDGAR EFTS (no CIK / N/A), web search, and the private-company carve-out as of 2026-06-30. All findings unaudited per public-source limitation for a private company.

---

Phase D — Project & stress-test

Lens 11 → rNPV-style value frame + runway-to-catalyst (no EPS; +clinical / +private)

No EPS exists and none was modeled. The right frame is option value on the platform vs. the next value-inflection catalyst.

• rNPV is not yet computable — the lead asset has no disclosed target, indication, or peak-sales basis, so a peak-sales × PoS × discount calculation would be pure fabrication. n/a (deliberately not invented).
• The question that actually matters (+clinical runway test): does cash reach the next value-inflection catalyst? The next catalyst is the 2027 internal IND / first-in-human. Terray's $120M (2024) + non-dilutive partner milestones plausibly fund IND-enabling work, but a clean clinic entry will likely require a Series C ``. The binary that gates everything: an internal molecule entering the clinic on the (already-slipped) 2027 timeline.
• Valuation sanity vs. peers ``: at ~$600M (2024 post), Terray is priced as a premium pre-clinical platform — cheaper than Isomorphic ($2.1B-round) and below a now-public Insilico, but richer than the data alone justifies unless the internal pipeline delivers. A flat-to-modest-up Series C would be a good outcome for current holders; a down round is the live risk if the 2027 clinic date slips again into a tougher capital market.
• No Brier forecast logged — unattended watchlist run; no committed base case for a pre-revenue private. (If forced to log one binary, it would be "Terray dosed first patient in an internal program by 31-Dec-2027," p≈0.45 `` given the one-year slip already on record.)

Lens 12 · Bull vs Bear

Bull case. Terray owns the scarcest input in small-molecule AI: a 13-billion-measurement, internally-generated, standardized binding dataset ~50× all public chemistry data, doubling annually — exactly the "garbage in" problem everyone else has, solved structurally. A peer-reviewed foundation model (COATI) plus a potency model that (if the claims hold) screens 26× more chemical space per GPU-hour means Terray can attack undruggable / novel-pocket targets where sparse-data rivals fail. NVIDIA is supplier + investor + distribution. Four pharma partnerships (BMS milestone already hit) provide non-dilutive validation cash, and the board (Xencor's Dahiyat, ex-Genentech Young, ex-Amgen Levy) can actually shepherd molecules through development. If even one internal immunology asset reaches the clinic and reads out, Terray re-rates from platform multiple to drug multiple — the Insilico playbook.

Bear case (2-3 permanent-impairment risks).

1. Discovery ≠ drugs. The entire moat sits upstream of the 90% of attrition that happens in humans. A bigger dataset makes you faster to a candidate; it does nothing for Phase 2 efficacy. DSP-1181 is the cautionary corpse. Zero molecules in humans + a clinic date that already slipped 2026→2027 is the core bear fact.
2. Value capture is downstream, where Terray is the weak party. It is one of 530+ AI-discovery shops; pharma holds the targets, capital, and commercialization. Partner economics (undisclosed, likely modest upfronts — Gilead ≈ $0.01 EPS hit) cap the partnered upside, and large pharma remains risk-averse to new platforms (AI deals carry only ~3% upfront vs ~6% for small-molecule asset deals ).
3. Unverified claims + capital-cycle risk. The most impressive metrics (TerraBind +20%/26×) are un-peer-reviewed and un-benchmarked. At a $600M pre-clinical post-money with no crossover fund on the cap table, a Series C is needed around the slipped clinic date — a down-round / dilution risk if 2027 slips again into a harsher market.

Pre-mortem (18 months out, thesis broke): It's end-2027. The internal IND slipped again (or a Phase 1 tox signal stalled the lead). Partners renewed nothing material; the BMS milestone was the high-water mark. A Series C closes flat-to-down because public AI-bio comps (Recursion-style multiple compression) reset private marks, and with no crossover money already in, the IPO window stays shut. The 13B-measurement dataset is still real — and still hasn't produced an approved drug, like everyone else's.

Are multiples too high? For a pre-clinical platform, $600M is full — justified only if you underwrite the internal pipeline working, not the data alone.

Contrarian view (what the market refuses to see): The bear consensus is "AI drug discovery hasn't produced an approved drug, so it's hype." The contrarian read: the data layer is quietly consolidating into a winner-take-most, and Terray's owned-microarray flywheel is one of maybe three credible private data monopolies (vs. Isomorphic's structure bet and Recursion's phenomics). If small-molecule AI value accrues to whoever owns the cleanest proprietary data at scale, Terray is mis-priced cheap — but only on a 3-5 year horizon and only if it survives the capital cycle to get there.

Lens 13 · Devil's Advocate (short-seller)

Dismantling the bull case:

• The moat monetizes the wrong stage. Proprietary binding data optimizes hit-to-lead, the cheapest, fastest, least-risky part of drug R&D. The expensive, value-determining failures are in vivo efficacy and human safety, which no amount of microarray data predicts. The moat is real and largely irrelevant to the outcome that matters.
• Concentration & dependency. Revenue (such as it is) is partner-milestone concentrated, undisclosed, and terminable — partners run multiple AI vendors and owe Terray nothing past contracted milestones. Compute is single-sourced to NVIDIA, which is also an investor (alignment today, leverage tomorrow).
• The most dangerous competitor bulls underestimate: not Isomorphic (also pre-clinical) — it's Schrödinger, the other chemistry-first platform, which has 1,600+ pharma users, a physics engine, AND internal assets already in Phase 1. Schrödinger can do data + physics + pipeline at public-company scale; Terray is data-only and pre-clinical.
• Worst capital-allocation / governance flags: none egregious, but brother co-founder concentration + a $600M mark with no crossover validation + un-peer-reviewed hero metrics is a stack of yellow flags that says "great science, unproven as a drug company."
• What must hold for the price: that the internal pipeline produces a clinical-stage asset on roughly the 2027 timeline AND that the data edge persists (it's doubling — but so is everyone's, and public datasets/AlphaFold-class tools keep raising the floor). Haircut growth/timeline by 20-30% (another year's slip, no internal IND until 2028) and the $600M mark is underwater at the next raise.
• Single permanently-impairing scenario: the lead internal asset fails in Phase 1 (tox or no target engagement) with no partnered asset advanced to compensate — the platform's first clinical test fails publicly, the "de novo discovery" narrative breaks, and the company is repriced as a services/data vendor, not a drug developer. Plausibility: moderate — it's a single-asset clinical bet with no human data yet.

Lens 14 · Management Questions (15, ordered by information value)

1. What is the specific target and indication of your lead internal program, and what is the exact, dated IND timeline — and why did the clinic date move from 2026 to 2027?
2. Has any independent third party benchmarked TerraBind's "+20% over Boltz-2 / 26× efficiency"? Will you publish a peer-reviewed, prospective validation?
3. What is your current cash runway in months, and does it reach a first-in-human read-out without a Series C?
4. For the internal lead, what specific evidence beyond binding affinity supports in vivo efficacy — the stage where AI-discovery has historically failed?
5. Across BMS/Gilead/Calico/Odyssey, how many targets are active, how many milestones earned to date, and what aggregate non-dilutive cash have they delivered?
6. What are the economic terms (upfront / milestone / royalty ranges) of the partnered deals, and what % of program value does Terray retain?
7. How dependent are you on NVIDIA DGX Cloud, and what is the cost/continuity/competitive risk of that single-vendor compute relationship given NVIDIA is also a shareholder?
8. Why is there no crossover fund on the cap table yet, and what milestones do you believe unlock a Series C — and an eventual S-1?
9. Your dataset is "doubling annually" — but as AlphaFold-class tools and public datasets improve, how do you keep the relative data edge from eroding?
10. What is your moat against Schrödinger, which has physics + the largest pharma install base + assets already in Phase 1?
11. How do you prioritize internal vs. partnered programs when the platform is the shared bottleneck, and which gets the best chemistry?
12. What is the insider/founder ownership structure, and how is the brother-co-founder concentration governed?
13. What would make you kill the lead internal program, and what is the go/no-go gate before IND spend accelerates?
14. What is your realistic base-rate assumption for your own programs' Phase 1 → approval probability, and how does it differ (if at all) from industry small-molecule base rates?
15. If the public AI-bio multiple stays compressed (Recursion-style) through your next raise, what is the financing plan that avoids a down round?

---