Scientific Due Diligence: Ditto Biosciences

Ditto Bio (YC W26) contends that parasites -- viruses, helminths, ticks -- have spent millions of years evolving proteins that suppress, redirect, or evade the human immune system, and that these proteins constitute an underexploited library of drug candidates for autoimmune disease. Rather than designing immunomodulators from scratch (the standard pharma approach, with a \~90% clinical-trial failure rate), Ditto starts from evolution-validated molecules and uses an AI platform called MoleculeMapper to systematically identify, characterize, and engineer them into therapeutics.

The logic is straightforward: if a parasite protein already controls a human immune pathway in vivo, much of the hard pharmacology has already been solved by natural selection. The remaining work is engineering -- improving stability, reducing immunogenicity, formulating for delivery.

Below is a rigorous assessment of the underlying science, the team's credibility, the computational platform, the competitive landscape, and the key risks.

1. The Biological Foundation: Is the Core Science Real?

Short answer: yes, with important caveats.

The Hygiene Hypothesis (Epidemiological Anchor)

The intellectual origin of this field is the "hygiene hypothesis" -- the observation that autoimmune diseases (IBD, MS, type 1 diabetes, asthma) have risen dramatically in industrialized nations where helminth infections have been eliminated, while remaining rare in regions where such infections are endemic. A landmark 2017 Nature Reviews Immunology paper confirmed that "the initial application of the hygiene hypothesis for autoimmune diseases proposed in the early 2000s has been confirmed and consolidated by a wealth of published data in both animal models and human autoimmune conditions." The inverse correlation between helminth exposure and autoimmune prevalence is one of the most robust findings in modern immunology.

The mechanistic explanation centers on regulatory T cells (Tregs). Helminths induce potent Treg expansion and IL-10/TGF-beta secretion to protect themselves from immune attack. This same regulatory cascade suppresses the Th1/Th17 responses that drive autoimmune tissue destruction. When you remove the parasites, you lose the Treg-inducing signal -- and in genetically susceptible individuals, the immune system tilts toward autoimmunity.

Helminth-Derived Proteins (Strong Preclinical Data)

The literature on defined helminth proteins as immunomodulators is substantial and growing:

• ES-62 (from Acanthocheilonema viteae): suppresses collagen-induced arthritis, lupus nephritis, and allergic airway inflammation in mice via IL-10 and inhibition of MyD88 signaling. The most well-characterized helminth immunomodulator.
• SJMHE1 (from Schistosoma japonicum): a 2025 Dove Press study demonstrated efficacy in a psoriasis model when delivered via hydrogel, showing this approach can work for defined peptides.
• A 2024 BMC Immunology systematic review of helminth-derived proteins in colitis models confirmed that multiple distinct proteins regulate NF-kB, TLR, and MAPK pathways -- suggesting this is not a single-protein curiosity but a broad class of evolved immunomodulators.
• A March 2026 Nature Scientific Reports meta-analysis further validated the approach across in vitro inflammation models, strengthening the case that helminth proteins represent a pharmacologically tractable class.

Tick Salivary Proteins (Strong and Underexplored)

Ticks must feed on their host for days to weeks without triggering immune rejection -- an extraordinary feat that requires sophisticated immunosuppression. The tick saliva literature is rich:

• Salp15 (from Ixodes scapularis): binds CD4 on T cells, inhibiting TCR signaling. A 2019 study in Frontiers in Immunology showed Salp15 achieves long-lasting CD4+ T cell suppression in a murine heart transplant model. This is a genuine in vivo proof-of-concept for tick-derived immunosuppressants.
• Tick-PD-1/PD-L1 pathway: A 2021 Nature Scientific Reports paper showed tick saliva induces PD-1 and PD-L1 expression on host cells -- mimicking the exact checkpoint pathway exploited by cancer immunotherapy (but in reverse, for immunosuppression). This is mechanistically elegant and clinically relevant.
• A 2020 Frontiers in Immunology review catalogued dozens of tick salivary proteins targeting complement, chemokines, histamine, bradykinin, and prostaglandins -- a deep well of evolved immunomodulators.

Viral Molecular Mimicry (Ditto's Highlighted Example)

Ditto's blog specifically analyzes viral molecular mimicry -- viruses that have evolved structural mimics of human immune proteins. A major October 2024 Nature Communications paper systematically catalogued molecular mimicry across all known human-infecting viruses, finding that structural mimics are "common" and concentrated in immune-regulatory pathways. The blog reportedly identifies a viral protein that mirrors the mechanism of a blockbuster autoimmune drug (likely a TNF or IL-6 pathway mimic, based on the autoimmune context) -- if validated experimentally, this is a compelling proof that evolution has independently converged on the same solutions pharma has found through decades of R&D.

Bottom line on the science: The biological foundation is not speculative. It rests on 20+ years of mechanistic immunology, robust epidemiological data, and increasingly well-characterized individual proteins. The field has moved well beyond hypothesis into defined molecular mechanisms.

2. The Cautionary Precedent: TSO Clinical Failure

Any diligence on this space must confront the elephant in the room: the failure of Trichuris suis ova (TSO) therapy in clinical trials.

Coronado Biosciences ran Phase 2 RCTs of live pig whipworm eggs in Crohn's disease and ulcerative colitis in the mid-2010s. The trials failed to show efficacy. This was a significant setback for the "helminth therapy" concept and soured many investors on the space.

However, the TSO failure is actually a validation of Ditto's differentiated approach, not a refutation:

• TSO used whole, live organisms -- an extraordinarily blunt instrument. A living worm secretes hundreds of proteins, many of which have nothing to do with immunomodulation (structural proteins, metabolic enzymes, reproductive proteins). Dosing live organisms introduces massive variability in immune exposure.
• The failure was in drug design, not target biology. Patients' immune responses to TSO were heterogeneous and unpredictable. A 2018 Frontiers in Pharmacology study found that NOD2 genotype variants significantly influenced treatment response -- suggesting the underlying biology is real but requires precision targeting, not a shotgun approach.
• Ditto's model is the opposite of TSO. They isolate specific, defined proteins with known or predicted targets, then engineer them as conventional biologics. This is the difference between eating willow bark (variable salicin content, GI toxicity) and taking aspirin (pure, dosed compound).

ParaTech ApS in Denmark still holds clinical-trial licenses for TSO, but the field has broadly moved away from live-organism therapy toward defined molecular approaches -- exactly where Ditto sits.

3. The Team: Credibility Assessment

Adair Borges, CSO (UCSF PhD)

Borges holds a PhD in Biomedical Sciences from UCSF (2014-2020) under Joe Bondy-Denomy, one of the world's leading CRISPR-Cas biology researchers. Her publication record shows \~30 papers with 58+ citations on Google Scholar, focused on:

• CRISPR-Cas systems and phage-bacteria interactions (her PhD work)
• Host-pathogen molecular interactions
• Anti-CRISPR proteins (which are themselves examples of pathogen-derived proteins that modulate host biology -- conceptually parallel to Ditto's thesis)

Post-PhD, she joined Arcadia Science as a founding scientist, where she co-built the ProteinCartography pipeline (published September 2023) and led the comparative phylogenomic analysis of chelicerates -- directly studying tick-derived proteins that suppress host detection. Her Arcadia work is essentially the prototype for Ditto's approach.

Assessment: Strong scientific pedigree. Not a parasitologist by original training (she's a molecular microbiologist), but her pivot from CRISPR-phage biology to parasite immunomodulators is intellectually coherent -- both domains involve understanding how foreign organisms manipulate host molecular machinery.

Dennis Sun, CEO (Berkeley PhD)

Sun holds a PhD from Berkeley in computational/evolutionary biology. Co-author on ProteinCartography and other Arcadia Science publications. His computational biology background complements Borges's wet-lab expertise.

Emily Weiss, CTO (UCSD PhD)

Weiss holds a PhD from UCSD, also formerly at Arcadia Science. Her background is in computational biology and AI-based protein structure prediction. She is the technical architect of the MoleculeMapper platform.

Team verdict: The three co-founders spent three years building tools at Arcadia Science that are directly ancestral to Ditto's platform. ProteinCartography (open-source, 43 GitHub stars) is the structural comparison engine; MoleculeMapper appears to be its evolution for therapeutic target identification. Combined \~40+ years of training at Harvard, Berkeley, UCSF, UCSD. The team is credible, domain-expert, and has a meaningful technical head start from their Arcadia work.

4. The Platform: MoleculeMapper and Computational Approach

What They Claim

MoleculeMapper ingests parasite proteomes (1M+ proteins analyzed to date across viruses, ticks, and worms), predicts which proteins interact with clinically validated human immune targets, and ranks candidates by drug-like properties (binding affinity, stability, solubility, predicted immunogenicity). They claim early experiments show 1-2 nM binding for surfaced candidates -- an impressive number if reproducible, since most approved biologics operate in the low-nM range.

What They've Published

Their blog details analyzing \~10,000 proteins from all known human-infecting viruses. They use structural comparison (likely via Foldseek + ProteinCartography lineage tools) and AlphaFold-Multimer for protein-protein interaction prediction.

The Critical Honest Signal: Arcadia's Self-Assessment

This is where the diligence gets interesting. In a 2025 publication from Arcadia Science -- authored by the same team -- they explicitly asked: "How confident should we be in potential targets of tick protease inhibitors predicted by AlphaFold-Multimer?" Their case study found that AF-Multimer predictions included:

• Plausible hits: proteases present in skin that connect to itch and inflammation (exactly what you'd expect tick salivary proteins to target).
• Non-physiologically relevant hits: human digestive enzymes and other proteins that a tick salivary protein would never encounter in vivo.

They concluded with genuine uncertainty about how to interpret their computational results, sharing the work openly to solicit community feedback.

This is simultaneously a risk flag and a credibility signal. Risk, because it shows the computational predictions have a high false-positive rate. Credibility, because the team is intellectually honest about this limitation rather than hiding it. It also explains why Ditto emphasizes experimental validation (the tissue biobank, binding assays) alongside computational prediction -- they know the AI alone is insufficient.

AlphaFold-Multimer Limitations (Structural Risk)

AlphaFold-Multimer has demonstrated strong performance for stable protein complexes where co-evolutionary signal exists in sequence databases. But parasite-host interactions are evolutionarily adversarial -- the parasite protein is under selection to bind the host target, but there's no co-evolutionary trace in the way that, say, a ribosomal complex has. A June 2025 bioRxiv preprint (Baptista et al.) from the Beltrao lab specifically studied AlphaFold's ability to model molecular mimicry in host-pathogen interactions, suggesting the field is still working out the limits.

Bottom line on the platform: MoleculeMapper is a sophisticated funnel, not an oracle. Its value is in prioritization: narrowing millions of proteins to thousands of candidates for experimental screening. The AI doesn't replace wet-lab validation, and the team appears to understand this.

5. Competitive Landscape

Ditto's differentiation is real: no other company combines (a) systematic computational mining of parasite proteomes with (b) AI-guided target identification at scale with (c) a tissue biobank for immunogenicity prediction. The closest analog in approach -- academic labs studying individual helminth proteins -- operates protein-by-protein rather than proteome-wide.

6. Key Scientific Risks

Risk 1: Translation Gap (High)

Parasite proteins evolved to work in the context of a live infection -- alongside dozens or hundreds of other co-secreted molecules, in specific tissue microenvironments, at specific concentrations. An isolated protein, dosed as a biologic, may not recapitulate the in vivo effect. This is the fundamental bet.

Risk 2: Immunogenicity (High)

Foreign proteins trigger antibody responses. Parasite proteins are inherently non-self. Ditto's tissue biobank approach to predicting immunogenicity is smart, but engineering low-immunogenicity variants of parasite proteins without destroying their function is a major technical challenge. Every modification to reduce immunogenicity risks reducing binding or efficacy.

Risk 3: Computational False Positives (Medium)

As their own Arcadia paper demonstrates, AF-Multimer predictions include substantial noise. If the hit rate from computation to validated binder is low, the platform becomes an expensive screening funnel rather than a precision targeting tool. The 1-2 nM binding claim for early candidates is encouraging but needs broader validation.

Risk 4: Delivery (Medium)

Protein therapeutics face standard biologic delivery challenges: parenteral administration, cold-chain storage, manufacturing complexity. Nothing specific to Ditto's approach makes this harder or easier than other biologics.

Risk 5: Regulatory Path (Low-Medium)

Engineered proteins from parasite sources are novel, but the regulatory framework for protein biologics is well-established (FDA's BLA pathway). The regulatory risk is more about the novelty of the source (parasites) creating unfamiliar safety questions for reviewers.

7. What Would Increase Conviction

1. Publication of the MoleculeMapper methodology in a peer-reviewed journal, with benchmark data showing hit rates from computational prediction to experimental validation.
2. In vivo efficacy data for at least one parasite-derived protein in an autoimmune animal model (colitis, EAE, or CIA would be standard).
3. Head-to-head comparison of a Ditto candidate against an existing biologic (e.g., anti-TNF) on the same target, demonstrating comparable or superior potency from the evolution-derived molecule.
4. Immunogenicity data from their tissue biobank showing they can predict and engineer around anti-drug antibody responses.
5. Expansion of the 1-2 nM binding claim across multiple targets with orthogonal validation (SPR, ITC, cell-based assays).

8. Overall Scientific Assessment

The science is real, the approach is differentiated, and the team is credible -- but the company is pre-IND and pre-publication, meaning the platform thesis is largely unproven outside of internal data.

The biological foundation (parasite-derived immunomodulators) rests on two decades of peer-reviewed evidence and is one of the more intellectually compelling angles in autoimmune drug discovery. The TSO clinical failure, often cited as a rebuttal, actually strengthens the case for Ditto's defined-molecule approach vs. whole-organism therapy.

The team's Arcadia Science pedigree is a genuine asset: they didn't just read about computational parasitology, they built the foundational tools (ProteinCartography) and honestly published the limitations (AF-Multimer false positives). The 7-month timeline from founding to 1M+ proteins analyzed and early binding data suggests rapid execution.

The key question is whether computational prioritization plus protein engineering can reliably convert parasite proteins into clinical drug candidates. This is an engineering challenge, not a fundamental science risk. The biology is validated; the translation is not. That's a better starting position than most early-stage biotechs, which are betting on unvalidated biology and unproven translation.

For a YC-stage company (\~$500K seed, <10 people, <1 year old), the risk-reward profile is strong. They need to publish, generate in vivo data, and demonstrate that MoleculeMapper's hit rate justifies the platform thesis. If they can show even a 10-20% hit rate from computational prediction to validated binder, the economics of mining 1M+ proteins become very attractive.