Proceedings · Session S-537 · filed November 20, 2026
AI & Emerging Tech in R&DSession paper
Tesorai raises $5.3 million to scale AI proteomics analysis
Tesorai raised $5.3 million, bringing total funding to $8 million, as the Verily-founded startup reports peptide gains, expands beyond proteomics and reanalyses five Alzheimer’s cohorts.
By Priya Raman4 min read897 words
Summary
- $5.3 million seed round raises Tesorai’s total funding to $8 million.
- Founded in 2024, Tesorai is led by former Verily colleagues Melissa Miao, Peter Cimermancic and Charlie Kim.
- Tesorai Search reports average unique-peptide gains of 12–68% against four named search engines.
- Tesorai Chat scored above 92% on a 94-case proteomics benchmark.
- A September 2026 preprint found p-tau217 in four of five reprocessed Alzheimer’s datasets.

By Priya Raman
Tesorai announced a $5.3 million seed round this week, lifting total funding to $8 million for an AI proteomics platform.
The round included Perplexity Fund, IQT, South Central Ventures, New Stack Ventures, NextGen Venture Partners, f7 Ventures, VitalStage Ventures and ZAKA. Representatives also said Tesorai had signed contracts with biopharmaceutical and research organizations.
Named collaborators include teams at MD Anderson Cancer Center, New York University and the Michael J. Fox Foundation. Tesorai did not disclose contract values, customer counts or revenue.
What does Tesorai sell?
Founded in 2024 by long-time Verily colleagues Melissa Miao, Peter Cimermancic, PhD, and Charlie Kim, PhD, Tesorai assigns them the roles of CEO, CTO and CSO. At Verily, Alphabet’s life-science company, Miao led business initiatives and strategic partnerships, Cimermancic led AI teams, and Kim launched the molecular science program.
Miao said: “That’s what inspired us to leave and build our own team specifically focused on building AI models to maximize the insights that you could extract from those data sets.”
Tesorai’s proprietary model underpins a platform that processes raw mass-spectrometry data from new laboratory experiments, public repositories and company archives. Kim said: “We can finally see the whole proteome in the way that genomics was able to 20 years ago.” He described mass-spectrometry analysis as difficult and AI as a way to extract more from each sample.
Tesorai plans integrative analysis across biological layers to build mechanistic understanding, starting with transcriptomics and later genomics. Its agentic interface accepts natural-language requests and processes large data volumes; funding supports expansion among biopharmaceutical and research customers, plus life-science partnerships.
Miao said applications span most drug-development stages, from target discovery through biomarker discovery and later development; Tesorai is also entering agrolife sciences.
What do the benchmarks show?
- Tesorai Search: A deep-learning engine described in a Journal of Molecular Biology paper reads full spectra and finds more peptides. Tesorai says its model resolves most of the 20 to 40 settings used by standard analysis tools.
The paper reports average gains of 12–68% in unique peptides against MaxQuant, FragPipe with MSBooster, PEAKS and Proteome Discoverer, and strong performance on data unseen during training.
Tesorai disclosed no dataset counts, variance estimates, benchmark composition, scoring method or independent replication for those comparisons, so the two headline performance figures remain company-reported rather than independently established.
-
Tesorai Trove: This catalog curates public human-tissue proteomics datasets and structures metadata into searchable tables for researchers and AI. Unlike tools limited to condensed literature datasets, Trove lets scientists mine raw files.
-
Tesorai Chat: Conversational prompts run proteomics quality checks, statistics and charts. On a 94-case proteomics benchmark whose results Tesorai shared in 2026, it exceeded 92% accuracy and ranked ahead of Phylo and GPT-5.4 among the tools tested.
Instead of training models on the fly, which can overfit and produce inaccuracies, Tesorai pre-trains on public repositories and other sources. Cimermancic said this supports generalization across experimental conditions and instrument types. He argued that heterogeneous data leave the model with no necessary biases, although he provided no bias metric.
Tesorai validates models trained on one manufacturer’s instruments using instruments from other manufacturers, according to the company. Academic partners have also designed experimental evaluations, including single-gene-knockout detection in cell lines; no outcomes were disclosed.
Tesorai has invested in scalable engineering and infrastructure to reprocess large repositories as instruments generate more data; Miao said software had not kept pace. It received its first platform patent in 2025. Company-shared data show more peptide and protein identifications and improved accuracy.
Customers generally buy tokens for usage rather than annual licenses, a new proteomics software model that Miao called more accessible; more free credits are planned for academic teams. Biopharma clients may buy service engagements that run in their cloud or support analysis sharing across departments.
Can five Alzheimer’s cohorts test the platform?
Kim cited a test of a newly published Alzheimer’s multiprotein plasma panel against data from five existing brain studies. The panel, described in a 2026 JAMA Neurology paper, could distinguish early- from late-stage Alzheimer’s disease.
Details appeared in the September 2026 bioRxiv preprint Rapid Mechanistic Bridging of an Alzheimer’s Disease Plasma Protein Staging Panel Across Brain Proteomics Cohorts. The team asked whether proposed blood proteins also tracked disease severity in brain proteomics.
Original protein summaries omitted p-tau217, an established plasma biomarker for amyloid-beta pathology in people at risk, and amyloid-beta 1-40 (Aβ40) from all five studies. Tesorai reprocessed raw spectra to quantify both proteoforms; p-tau217 then appeared in four of five datasets despite its absence from the summaries.
The preprint reports that the panel distinguished early from late disease across cohorts, with other proteins showing the same pattern, and supported a mechanistic account of blood-protein changes. Tesorai completed the reanalysis in three weeks without running a new study, illustrating its claim that raw files can answer questions original investigators did not pose.
The interview disclosed no cohort sample sizes or preprint funding, and no prospective patient-outcome test was conducted. The work therefore shows retrospective data reuse, not clinical utility or prospective performance.
The next test for Tesorai’s $5.3 million investment is whether independent academic and industrial users reproduce the reported peptide gains across instruments and planned omics datasets.
via sciencedirect.com (Original)
Filed under
- ai-proteomics
- mass-spectrometry
- biotechnology-funding
- drug-discovery
- alzheimer-s-biomarkers
More from Priya Raman
Show full bio
Staff writer covering money, funding and market shifts around research and development at Hypothesis Wire.
289 articles
References
- Pharma's $29.5 Billion AI Bet Faces a Seven-Year Wait for Proof
- Isomorphic Labs banks $5 billion in biotech's largest venture round
- Isomorphic Labs predicts AI-designed drugs as $5bn raise talks surface
- Pfizer to Deploy Edison's Kosmos AI Scientist Across R&D Pipeline
- Anthropic and Novo Nordisk expand Claude work into drug discovery