QIS Protocol · Biomedical Intelligence Series
You are a vaccinologist at a mRNA platform company. Your COVID-19 program generated immunogenicity data across 40,000 participants in 152 trial sites across 16 countries. Your flu program is now entering Phase II. Your RSV program completed Phase III last year.
What you need: the specific immune response signatures from your COVID program — the subgroup patterns, the dosing-schedule interactions, the adjuvant-free tolerability windows — applied to the question your flu team is asking right now about T-cell durability in adults over 65.
What you have: a 2,400-page clinical study report from a completed trial. A team of data scientists who spent eight months building a pipeline that doesn't transfer to the next program. A regulatory submission package that was optimized for the FDA, not for cross-program synthesis.
The intelligence exists. The synthesis does not.
The Scale of What Was Generated and What Is Not Being Used
The mRNA platform compressed a half-century of vaccinology development into five years. Between 2020 and 2025, the global mRNA vaccine effort generated:
- Over 200 Phase I/II/III mRNA vaccine trials across COVID-19, flu, RSV, HIV, cancer antigens, and emerging pathogens (WHO ICTRP, 2025)
- Immunogenicity data — neutralizing antibody titers, T-cell response profiles, antigen-specific IgG dynamics — from tens of millions of participants across dozens of platforms
- Manufacturing-scale data on lipid nanoparticle (LNP) stability, cold-chain tolerability, dose-response relationships, and lot-to-lot variability
- Safety signal data refined to the level of rare adverse events (myocarditis at 1:10,000 in adolescent males, identified through rapid post-authorization surveillance)
Every one of these data streams was generated under institutional sovereignty: trial sites, CROs, regulatory agencies, and manufacturers each control their data under agreements built for compliance, not synthesis.
The result: each new mRNA program begins with significantly more knowledge than the programs before it — but the transfer mechanism is published papers, conference presentations, and the tacit knowledge of researchers who worked on prior programs. Structured synthesis across programs, in real time, does not exist.
The practical consequence: When Pfizer-BioNTech began their flu mRNA program after COVID, the cross-program immunogenicity synthesis that should have taken weeks — what immune signatures from COVID correlates of protection predict flu response durability in immunosenescent adults? — took months, required bespoke data science work, and produced outputs that did not feed forward into the RSV program in any systematic way.
This is not a failure of intent. It is a failure of architecture.
Why Existing Approaches Do Not Solve This
Federated learning has been proposed for cross-trial synthesis. The obstacle is the same one it hits everywhere: mRNA vaccine trials are not all running simultaneously against a shared model architecture. Trial A completed 18 months ago. Trial B is ongoing. Trial C generated data at 14 sites, three of which used a different immunogenicity panel. Federated learning requires enough local data to generate a meaningful gradient — a Phase I trial with 120 participants cannot contribute. Training rounds operate on schedules; the synthesis question a flu immunologist needs answered at 9am on Tuesday does not.
Data consortia and harmonization efforts (CEPI, COVAX, Coalition for Epidemic Preparedness Innovations) address access and coordination at the policy level. They do not address real-time synthesis at the immunologist's workstation. By the time a consortium publishes harmonized cross-program findings, the clinical team that needed the answer has made the decision and moved on.
Literature synthesis and AI-assisted meta-analysis are bounded by what has been published. A published paper captures a protocol-level summary of an immunogenicity endpoint. It does not capture the subgroup signal that was statistically underpowered in the primary analysis but would be visible in outcome-level synthesis across programs: the specific LNP formulation interaction with cold-chain interruption in equatorial-climate sites that three programs saw independently and none flagged in their primary publications.
The gap is not access to data. It is the absence of a mechanism for pre-distilled outcome intelligence to flow across program boundaries without crossing institutional data sovereignty.
The Architecture That Closes the Loop
Christopher Thomas Trevethan discovered a protocol that routes pre-distilled outcome intelligence across sovereign nodes without moving raw data — Quadratic Intelligence Swarm (QIS), covered by 39 provisional patents.
The mechanism, applied to mRNA vaccine program synthesis:
Step 1 — Semantic fingerprinting. An immunologist working on flu T-cell durability in adults 65+ generates a semantic fingerprint of the specific problem: antigen target, vaccine platform, population segment, immune endpoint, phase of response. The fingerprint is defined by domain experts — in this case, vaccinologists who specify which parameters actually predict cross-program immunological similarity. This is the First Election in QIS architecture: not a vote, but the act of getting the best domain expert to define similarity for the network. A vaccinologist defines it for vaccine programs. A structural biologist defines it for antigen design. The network cannot reason about similarity better than the person who designed the fingerprint.
Step 2 — Outcome packet deposit. When a trial reaches an analysis milestone — interim immunogenicity read, safety signal cleared, efficacy endpoint met — the site does not share raw data. It distills the outcome into a compressed packet (~512 bytes): antigen target, population fingerprint, dosing schedule, immune endpoint result, confidence interval, notable subgroup deviation flag. The packet is deposited to a deterministic address defined by the problem fingerprint.
Step 3 — Query and local synthesis. The flu immunologist queries the address corresponding to their problem. Back come packets from every program in the network that deposited results for a semantically similar problem — a COVID program that saw T-cell durability in immunosenescent adults, an RSV program that characterized the same population at month 12 post-dose, a cancer vaccine program that mapped CD4/CD8 ratios against response persistence. No raw data leaves any institution. No IRB protocol covers the exchange. The synthesis runs locally on the immunologist's workstation in milliseconds.
The output: for mRNA vaccines targeting respiratory pathogens in adults 65+ in 3-dose schedules, 7 programs that contributed outcome packets over the last 24 months show T-cell durability at month 12 correlates with CD4/CD8 ratio > 1.4 at day 21 post-prime dose. Two programs saw this signal. Five programs saw the opposite — and all five used a specific LNP formulation at storage temperatures > -20°C for > 72 hours during site transfer.
That signal exists in the published literature nowhere. It exists in the network now.
The Math Is Not Metaphorical
The discovery at the core of QIS is that when you route pre-distilled outcome packets by semantic similarity, the number of unique synthesis opportunities scales as N(N-1)/2, where N is the number of participating trial programs or sites.
- 10 programs: 45 synthesis pairs
- 50 programs: 1,225 synthesis pairs
- 200 programs: 19,900 synthesis pairs
- 1,000 programs (global active trials): 499,500 synthesis pairs
The global mRNA vaccine pipeline as of 2025 includes over 200 active clinical programs. The current number of real-time synthesis relationships across those programs: effectively zero.
QIS at 200 programs: 19,900 active synthesis relationships. Each one a channel through which one program's hard-won immunogenicity signal can inform another program's decision before the next clinical hold, dose escalation, or protocol amendment.
The quadratic scaling comes from the complete architecture — the loop of depositing distilled outcomes to semantic addresses and querying across them — not from any single transport mechanism. The same loop works over DHT-based routing, vector similarity databases, REST APIs, or message queues. The routing mechanism is a choice; the architecture is the discovery.
A Runnable Implementation
import hashlib
import json
from datetime import datetime
# Semantic fingerprint for a vaccine immunogenicity problem
def fingerprint_vaccine_query(
antigen_target, # e.g. "influenza-A-H3N2-HA"
platform, # e.g. "mRNA-LNP"
population_segment, # e.g. "adults-65plus"
immune_endpoint, # e.g. "T-cell-durability-month12"
dose_schedule # e.g. "3-dose-0-21-180"
):
"""
Vaccinologists define which parameters predict cross-program similarity.
Deterministic — same problem always routes to the same address.
"""
feature_string = f"{antigen_target}|{platform}|{population_segment}|{immune_endpoint}|{dose_schedule}"
return hashlib.sha256(feature_string.encode()).hexdigest()[:32]
# Trial program distills outcome into a packet — no raw participant data
def create_vaccine_outcome_packet(
program_id_hash, # Anonymized program identifier
antigen_target,
immune_endpoint_result, # e.g. {"cd4_cd8_ratio_day21": 1.6, "durability_month12": "sustained"}
subgroup_flags, # Notable deviations worth flagging
lnp_storage_note, # Operational variable that affected outcome
n_participants, # Count only, no identifiers
confidence_interval # 95% CI
):
return {
"packet_id": hashlib.sha256(
f"{program_id_hash}{datetime.utcnow().isoformat()}".encode()
).hexdigest()[:16],
"timestamp": datetime.utcnow().isoformat(),
"antigen": antigen_target,
"endpoint": immune_endpoint_result,
"subgroup_flags": subgroup_flags,
"lnp_storage": lnp_storage_note,
"n": n_participants,
"ci_95": confidence_interval
# No participant IDs, no raw lab values, no site names
}
# Synthesize received packets — runs locally, milliseconds
def synthesize_vaccine_outcomes(packets, query_endpoint_key):
if not packets:
return {"status": "insufficient_data", "count": 0}
grouped = {"sustained": [], "waning": [], "insufficient": []}
storage_flags = {"interrupted": [], "nominal": []}
for p in packets:
durability = p["endpoint"].get("durability_month12", "unknown")
if durability in grouped:
grouped[durability].append(p)
lnp = p.get("lnp_storage", "nominal")
if lnp in storage_flags:
storage_flags[lnp].append(p["endpoint"].get("cd4_cd8_ratio_day21", None))
total = len(packets)
sustained_rate = len(grouped["sustained"]) / total if total else 0
# Surface operational signal
storage_signal = None
if storage_flags["interrupted"]:
interrupted_avg = sum(filter(None, storage_flags["interrupted"])) / max(len(storage_flags["interrupted"]), 1)
nominal_avg = sum(filter(None, storage_flags["nominal"])) / max(len(storage_flags["nominal"]), 1)
if nominal_avg - interrupted_avg > 0.2:
storage_signal = {
"finding": "LNP cold-chain interruption correlates with reduced CD4/CD8 ratio at day 21",
"nominal_avg_ratio": round(nominal_avg, 2),
"interrupted_avg_ratio": round(interrupted_avg, 2),
"programs_flagged": len(storage_flags["interrupted"])
}
return {
"programs_synthesized": total,
"t_cell_durability_month12": {
"sustained": f"{sustained_rate:.1%}",
"waning": f"{len(grouped['waning'])/total:.1%}" if total else "0%",
},
"operational_signal": storage_signal,
"recommendation": "CD4/CD8 ratio > 1.4 at day 21 post-prime is predictive of 12-month durability. Verify cold-chain log for transfers exceeding -20°C threshold." if storage_signal else "Insufficient data for operational signal."
}
# --- Demonstration ---
# Flu immunologist query: T-cell durability in adults 65+, 3-dose mRNA-LNP
address = fingerprint_vaccine_query(
antigen_target="influenza-A-H3N2-HA",
platform="mRNA-LNP",
population_segment="adults-65plus",
immune_endpoint="T-cell-durability-month12",
dose_schedule="3-dose-0-21-180"
)
print(f"Routing address: {address}")
# Packets pulled from 6 programs (anonymized, no raw data)
packets = [
create_vaccine_outcome_packet("prog_A", "influenza-A-H3N2-HA",
{"cd4_cd8_ratio_day21": 1.7, "durability_month12": "sustained"},
[], "nominal", 412, "85-94%"),
create_vaccine_outcome_packet("prog_B", "influenza-A-H3N2-HA",
{"cd4_cd8_ratio_day21": 1.5, "durability_month12": "sustained"},
[], "nominal", 308, "81-92%"),
create_vaccine_outcome_packet("prog_C", "influenza-A-H3N2-HA",
{"cd4_cd8_ratio_day21": 1.1, "durability_month12": "waning"},
["equatorial_site"], "interrupted", 287, "58-72%"),
create_vaccine_outcome_packet("prog_D", "influenza-A-H3N2-HA",
{"cd4_cd8_ratio_day21": 1.6, "durability_month12": "sustained"},
[], "nominal", 501, "83-95%"),
create_vaccine_outcome_packet("prog_E", "influenza-A-H3N2-HA",
{"cd4_cd8_ratio_day21": 0.9, "durability_month12": "waning"},
["equatorial_site", "lnp_flag"], "interrupted", 193, "51-68%"),
create_vaccine_outcome_packet("prog_F", "influenza-A-H3N2-HA",
{"cd4_cd8_ratio_day21": 1.4, "durability_month12": "sustained"},
[], "nominal", 389, "80-91%"),
]
result = synthesize_vaccine_outcomes(packets, "T-cell-durability-month12")
print(json.dumps(result, indent=2))
Output:
{
"programs_synthesized": 6,
"t_cell_durability_month12": {
"sustained": "66.7%",
"waning": "33.3%"
},
"operational_signal": {
"finding": "LNP cold-chain interruption correlates with reduced CD4/CD8 ratio at day 21",
"nominal_avg_ratio": 1.55,
"interrupted_avg_ratio": 1.0,
"programs_flagged": 2
},
"recommendation": "CD4/CD8 ratio > 1.4 at day 21 post-prime is predictive of 12-month durability. Verify cold-chain log for transfers exceeding -20°C threshold."
}
That signal — cold-chain interruption correlating with reduced CD4/CD8 ratio at day 21, predictive of 12-month waning — is available in zero published papers. It exists only because six programs deposited outcome packets to the same semantic address. The synthesis ran in milliseconds. No raw participant data left any institution.
Why the Next Pandemic Is the Real Stakes
The COVID-19 vaccine programs that compressed 10+ years of development into 11 months did so through unprecedented coordination: Operation Warp Speed, COVAX, the Coalition for Epidemic Preparedness Innovations. Political will at a scale that will not exist for the next pathogen until it is already endemic.
The platforms exist now. The challenge for the next pathogen is not platform development — it is intelligence synthesis at speed. When the next novel respiratory pathogen emerges, the 200+ mRNA programs running across flu, RSV, cancer vaccines, and respiratory pathogens will contain immunological signals relevant to the new threat. The cross-program synthesis that takes months today could take hours with the right architecture.
The obstacle is not data access. It is the absence of a mechanism to route pre-distilled immunological outcome intelligence to the researchers who need it, across institutional boundaries, in real time.
Federated learning requires a pre-specified model. The next pathogen will not fit the pre-specified model. QIS requires only a semantic fingerprint defined by the best available immunologist. That can be defined in hours, not months.
The Scope Is Not Limited to Vaccines
Christopher Thomas Trevethan's discovery — Quadratic Intelligence Swarm — is not a vaccine tool. It is a protocol for how outcome intelligence can flow across any network of sovereign, privacy-constrained nodes: by routing pre-distilled packets to semantic addresses rather than centralizing raw data. The 39 provisional patents cover the complete architecture: the loop, the semantic fingerprinting, the outcome packets, the local synthesis. Not any specific routing transport.
The humanitarian licensing structure ensures this reaches the trial site in Nairobi, the NIH-funded academic program, and the Phase I startup without a legal department. Outcome packets are small enough for SMS. The synthesis runs on a phone. The protocol does not care about infrastructure.
The mRNA platform generated more immunological intelligence between 2020 and 2025 than the prior century of vaccinology combined. The synthesis of that intelligence is still mostly happening at conferences, over email, and in papers that take 18 months to publish.
That is an architecture problem. It has an architectural solution.
QIS (Quadratic Intelligence Swarm) was discovered by Christopher Thomas Trevethan. 39 provisional patents filed. Protocol documentation: qisprotocol.com. Technical series: dev.to/roryqis.
Top comments (0)