Translational omics, first

Analyses that survive the manuscript, the IND, and the board.

When a computational-biology lead, PI, or founder needs an omics result they can defend — not a slide that collapses under review — Biosync owns the analysis end to end. One senior scientific owner. Reproducible methods. NDA before any files move.

See the first engagement Smriti — agent memory

1 owner

Senior scientist accountable from kickoff to handoff

Methods

Omics, clinical data, graphs — named in the SOW, not a logo wall

BAA

HIPAA-aligned · PHI stays in your cloud · SOC 2 controls on request

Public corpora we work with

These are data sources and reference collections — not clients, not logos, not endorsements. Proprietary clinical datasets are handled only after NDA (and a BAA when PHI is in scope).

NCBI GEO TCGA GTEx ENCODE HMP CELLxGENE Human Cell Atlas Tahoe-100M DepMap gnomAD NCBI SARS-CoV-2 MIMIC SEER CMS UK Biobank All of Us HAGR (GenAge / DrugAge) Aging Atlas

Standards

Built for rooms where methods are audited and figures get cited.

Compliance

HIPAA-aligned, BAA-ready

PHI engagements run under a signed BAA, inside your HIPAA-eligible cloud tenancy. We do not claim HIPAA certification.

Reproducibility

Audit-grade pipelines

Version-pinned, containerized workflows with full provenance — so internal review, a journal, or a regulator can rerun the work.

Data standards

CDISC & OMOP-aware

Clinical and real-world data handled with SDTM, ADaM, and OMOP CDM conventions when submission-grade evidence is the goal.

IP & publication

Your data, your science

Clear IP terms in every SOW. Publication-ready methods and co-authorship where the work merits it — standard for academic engagements.

Security posture, sample BAA, and senior-owner CV available under signed NDA. Request NDA.

Fit

Who this is for — and who should keep looking.

Better a short no than a mismatched sprint. If you need a wet lab, a regulatory-affairs owner, or a standing 20-person analytics bench, we are the wrong firm.

A fit

Write if you are

A computational-biology or translational lead who needs a defensible omics analysis without hiring a group.
An academic PI who needs publication-grade methods, figures, and a methods section that will survive review.
A digital-health founder who needs a public-data baseline before a pilot, a diligence memo, or a board conversation.
A team with data in hand and a bottleneck — MoA, atlas, biomarker, RWE, or a reasoning layer over the literature.

Not a fit

We do not

Run wet-lab assays, clone, or operate a core facility.
Own regulatory submissions (IND, NDA, 510(k)). We supply analysis the sponsor’s team can file.
Staff a large standing analytics org or act as a body shop.
Take PHI or unpublished data over the public form. NDA first; BAA when PHI is in scope.

First engagement

A 2–6 week sprint. One bottleneck. A written deliverable you can kill.

No retainer required to start. No dollar amounts on this page — the quote is tied to the deliverable we agree in writing. Smriti can be a parallel product conversation (discovery call, not a price).

Request a discovery call Send NDA first

01

NDA before files

Same inbox: hello@bio-sync.tech. We do not accept data, PHI, or unpublished sequences through the public form.
02

Fixed scope, explicit artifact

A report, pipeline, figures, or diligence memo — named in the SOW, with success criteria you can point at.
03

One senior owner

The scientist on the kickoff is the scientist on the midpoint and the handoff. CV after NDA.
04

Kill switch

If the data or the question does not hold, we stop. You keep what was produced to that point; we do not stretch a bad sprint.

Three lanes · omics is the door

We start where the biology is. The other two lanes exist so the same question can finish.

Most engagements begin as bulk, single-cell, spatial, or multi-omic analysis. Healthcare data and biomedical graphs come in when the same decision needs clinical context or a reasoning layer over literature and databases.

Primary

01 / Omics

Multi-omics & bioinformatics

Publication-grade analyses, built to be rerun, audited, and extended — not one-off notebooks.

Bulk RNA-seq and single-cell RNA-seq
Spatial transcriptomics and cell atlasing
Proteomics, alongside matched RNA
Epigenomics — ATAC-seq and ChIP-seq
Neoantigen / HLA / mRNA antigen selection (HLA typing → somatic variants → epitope / peptide–MHC → ranked antigens)
CRISPR-screen RNA-seq QC gates (depth, alignment, batch) before models
Multi-omic integration and biomarker ID

02 / Healthcare data

Clinical & real-world analytics

Public and proprietary clinical datasets turned into a baseline a hospital, payer, or investor can interrogate.

Critical-care analytics (MIMIC-III / IV)
Injury surveillance (NEISS)
Claims, eligibility & denials
Real-world evidence & RWD
Outcomes & risk stratification

03 / KG & AI

Biomedical knowledge graphs

A queryable layer over genes, diseases, drugs, and literature — so hypothesis generation is a path, not a pile of PDFs.

Heterogeneous GNN & link prediction on multi-omics KGs
Agentic pipeline, graph, and tool orchestration
Literature RAG with citations
Biomedical knowledge-graph construction
Target discovery & repurposing
Ontology harmonization & FAIR data

Antigen

Neoantigen / HLA / mRNA

HLA typing → somatic variants → epitope / peptide–MHC → ranked antigens for oncology (peptide or mRNA).

Screens

CRISPR-screen RNA-seq QC

Depth, alignment, and batch gates before any model is fit. Guide coverage and dropout after the gates pass.

Graphs

Heterogeneous GNN

Link prediction on multi-omics knowledge graphs — genes, drugs, diseases, assays.

Agents

Pipeline orchestration

Agentic wiring of pipeline, graph, and tools so a run is a path, not a pile of notebooks.

Literature

RAG with citations

Answers that name their papers. Uncited claims do not ship.

How we work

One path. Diagnose, scope, analyze, hand off.

Two overlapping “four-step” stories help no one. This is the engagement. The scientific stages sit underneath, for omics work, so you can see what actually happens to the data.

01 · Diagnose

Name the bottleneck

Before any code: the decision you need to defend, the constraint blocking it, and whether we are the right owner.

02 · Scope

A defensible first win

Fixed-scope sprint (2–6 weeks), written deliverable, success criteria, kill switch. NDA signed before files move.

03 · Analyze

Weekly working demos

You see the story as it forms — QC, models, figures — and steer it. Not a quarterly slide dump.

04 · Hand off

You own the narrative

Figures, methods, pipeline, and a runbook your team can rerun. The senior owner is on the final call.

Inside an omics engagement

Ingest

Raw intake, metadata harmonization, QC, batch-effect assessment, reproducibility setup.

Process

Alignment and mapping, quantification, normalization, peak or signal extraction, variant calling.

Analyze

Differential tests, clustering and cell typing, trajectory, motif and regulatory inference, pathways.

Deliver

Multi-omic integration, statistical or ML models, figures, dashboards, runbooks, methods text.

Representative work

Methods we ship — not a logo wall.

Problem shapes and analyses we have run. Named clients, full write-ups, and code samples sit behind NDA. Outcomes below are from the original public descriptions of this work; we are not adding new metrics here.

Filter by lane. Expand a card for problem, approach, and outcome.

Featured · Translational

Bulk RNA-seq · methods

Uncovering drug mechanism-of-action in tumor models

Lead compound worked in xenografts but the MoA was unclear — blocking IND-enabling work.

ProblemEfficacy signal without mechanism — regulatory narrative was stalled until the team could explain why it worked.

ApproachQC & alignment → differential expression (treated vs. control) → pathway enrichment → upstream regulator analysis → cross-reference with public drug-response databases.

OutcomeSurfaced two converging signaling pathways and a candidate resistance marker — anchored the MoA story for regulatory filings.

Differential expressionPathway analysisBiomarker ID

Featured · Immuno-oncology

Single-cell RNA-seq · methods

Building a single-cell atlas of the tumor immune microenvironment

47 patient samples, no unified view of which immune states drove response vs. resistance.

ProblemData existed but fragmented across batches; no cell-state framework to support patient stratification.

ApproachHarmonized QC across batches → integration & clustering → cell-type annotation → responder vs. non-responder differential testing → trajectory analysis on T-cell exhaustion.

OutcomeTwo previously unappreciated exhaustion-associated states identified — now anchor the lead program's patient stratification strategy.

IntegrationCell annotationTrajectory

Featured · Digital health

Healthcare analytics · MIMIC-III

Early sepsis prediction in the ICU

Investor diligence needed a defensible ICU early-warning baseline before clinical pilots.

ProblemTeam had a model idea but no rigorous benchmark to stand behind in conversations with hospitals and investors.

ApproachMIMIC-III cohort extraction → feature engineering on vitals, labs & meds → time-aware splits → benchmarking against Sepsis-3 → interpretability via SHAP.

OutcomeValidated predictive model with AUROC > 0.85 at 6 hours pre-onset — de-risked diligence and anchored the first hospital pilot.

Cohort extractionTime-series MLInterpretability

Featured · Drug discovery

Biomedical knowledge graph · methods

A biomedical knowledge graph for drug-repurposing hypothesis generation

Scanning literature and public databases for an orphan disease — at a scale humans can't.

ProblemManual literature review was unfeasible; scattered sources made systematic hypothesis generation impossible.

ApproachBuilt a KG linking genes, diseases, drugs, pathways & literature → entity resolution & ontology harmonization → graph embeddings → path-based scoring of drug-disease hypotheses → LLM-assisted evidence summarization.

OutcomeRanked list of 14 candidate molecules with mechanistic rationale — two progressed into in-vitro validation.

OntologyGraph embeddingsLLM + RAG

AI for biology · therapeutic antibodies

Antibody–antigen binding prediction for therapeutic screening

Predict binding from antibody sequence alone — compressing months of wet-lab screening into hours of compute, using only public datasets.

ProblemTherapeutic antibody discovery is bottlenecked by low-throughput binding assays. Teams need a computational pre-filter that prioritizes candidates before burning wet-lab capacity — on defensible, publicly reproducible data.

InputsAntibody heavy & light chain sequences (FASTA); optionally the antigen sequence or a pre-computed embedding for the target.

ApproachBaseline: engineered sequence features (CDR lengths, physicochemical descriptors, k-mer composition) feeding gradient-boosted models and random forests. Advanced: protein-language-model embeddings (ProtBERT, ESM-family) of paired heavy/light chains concatenated with antigen embeddings, fed into a neural classifier with calibration and uncertainty estimation.

DatasetsPublic only. SAbDab for positive pairs and structural priors, OAS for repertoire-scale sequence coverage, and CoV-AbDab for a focused viral-target use case. Negatives drawn via matched decoy sampling from OAS.

OutputCalibrated binding probability per (antibody, antigen) pair, with an interpretability layer surfacing which CDR regions drive the prediction.

OutcomeA defensible ML pre-filter that compresses the funnel and frees wet-lab capacity for the highest-probability candidates.

Protein LMs · ESM / ProtBERTSAbDab · OAS · CoV-AbDabCalibrated binding probabilityCDR interpretability

ChIP-seq · epigenomics

Mapping transcription-factor binding across a cancer cohort

How does a master TF rewire the chromatin landscape in resistant vs. sensitive cells?

ProblemThe lab had TF ChIP data but no integrated view linking binding to functional consequence.

ApproachAlignment → peak calling with matched input controls → differential binding → motif enrichment → integration with matched RNA-seq to link binding to expression.

OutcomeDelivered a prioritized shortlist of TF-regulated genes underpinning resistance — anchored a top-tier manuscript submission.

Peak callingMotif analysisChIP + RNA

ATAC-seq · developmental biology

Charting chromatin accessibility across a differentiation time course

Which regulatory elements open and close across a 7-day differentiation protocol?

ProblemProtocol worked inconsistently — team needed a regulatory map to understand which TFs drove key transitions.

ApproachQC & fragment-size assessment → alignment → peak calling → differential accessibility across timepoints → footprinting for TF activity → integration with matched expression.

OutcomeProduced a regulatory roadmap — identified a pair of TFs whose perturbation boosted protocol efficiency by ~40%.

AccessibilityFootprintingTime-course

Spatial transcriptomics · translational

Spatial mapping of tumor-immune interfaces

Why do some tumors respond while histologically similar ones don't?

ProblemBulk and single-cell data couldn't explain divergent clinical outcomes; spatial context was missing.

ApproachSpatial QC & deconvolution → tissue-region segmentation → neighborhood analysis of immune infiltration at tumor boundaries → correlation with clinical outcomes.

OutcomeIdentified a spatial immune-exclusion signature at invasive margins that predicted response — adopted into the biomarker hypothesis for an ongoing trial.

DeconvolutionNeighborhoodsClinical correlation

Multi-omic integration · academic

Fusing transcriptomics, methylation, and proteomics into one model

Three modalities, three silos, quarterly cycles. The integration problem no one owned.

ProblemEach modality was analyzed separately; insights never triangulated across them.

ApproachHarmonized sample metadata → per-modality normalization → factor-based integration → joint clustering → pathway & network inference.

OutcomeCompressed analysis turnaround from quarters to weeks; contributed core figures to two peer-reviewed publications.

Factor analysisJoint clusteringNetworks

Healthcare analytics · NEISS

Consumer-product injury surveillance for a policy team

A defensible national estimate of emerging injury trends across a 10-year window.

ProblemFree-text narratives made it impossible to track a specific product category with standard reports.

ApproachNEISS ingestion & harmonization → narrative NLP for product-specific injuries → weighted national estimates → demographic stratification → interactive surveillance dashboard.

OutcomeUncovered a 3.2× increase in a previously unflagged injury category — cited in a federal regulatory recommendation.

NLPWeighted estimationSurveillance

Public cancer data · SEER

Survival disparities across demographic cohorts

Rigorous evidence of survival differences for a cancer subtype — ready for peer review.

ProblemAdvocacy team needed hazard-ratio-level evidence, not descriptive statistics, to unlock grant funding.

ApproachSEER cohort construction → Kaplan–Meier → Cox PH modeling with adjustment for stage & treatment → sensitivity analyses.

OutcomePeer-review-ready report quantifying hazard ratios across cohorts — drove a grant-funded intervention program.

Survival analysisCox modelingHealth equity

Payer data · claims & denials

Denials triage & recovery engine for a Series A startup

80k+ denied claims per month. Manual review capping recovery at 18%.

ProblemVolume outstripped the RCM team; high-dollar denials were treated the same as trivial ones.

ApproachClaims ingestion (837/835) → reason-code classification → root-cause clustering → appeal prioritization by recoverable $ → HITL queue for high-dollar cases.

OutcomeLifted recovery rate by 31 points; compressed cash cycle by 11 days — paid back in its first quarter.

Claims parsingClassificationWorkflow

Public payer data · CMS

Readmission-risk modeling across Medicare cohorts

Which beneficiaries are most likely to be readmitted within 30 days?

ProblemCare-management team was overwhelmed; no triage logic to focus resources on the highest-risk discharges.

ApproachCMS claims cohort definition → feature engineering on diagnoses, procedures, prior utilization → gradient-boosted model with calibration → subgroup fairness audit.

OutcomeProduction-grade model with calibrated risk tiers — enabled care-management to triage 3× more patients per nurse.

Claims modelingCalibrationFairness

Knowledge graph over a DNA strand, assay plate, and tissue map

Product · Smriti

Smriti — agent memory you can defend.

For teams shipping agents into trials, drug safety, and translational research. Every output reconstructs from stored evidence. Every claim cites a source. Every change is verifiable. Self-hosted. Your machine.

Provenance

Where did the agent get this? The retrieval set is stored, not discarded. Compliance does not start from three log files.

Stale protocol

Was that guideline still current? Every link carries valid-from and valid-until. Superseded text cannot quietly resurface.

Contradictions

Did the agent see the conflict? Disagreement is a signal, not a silent overwrite.

Three primitives, wired into storage: a tamper-evident log, memory that earns its place, and citations the system enforces — rejected at write time, not logged after the fact.

Product page Open a live chain hello@bio-sync.tech

The practice

One senior scientific owner. CVs after NDA.

Who does the work

A small scientific practice, not a hidden bench.

Biosync is led by one senior scientific owner who is accountable — on paper — for the science and the deliverable. That person is on the kickoff, the midpoint, and the handoff. We do not farm analysis to unnamed juniors or silent subcontractors.

The roster is NDA-gated because most engagements sit under client confidentiality, not because the practice is anonymous. After NDA you get the owner’s CV, relevant prior work, and references. If a problem needs a specialist (clinical informatics, graph engineering, a particular assay type), that person is named under the same NDA.

NDA

CV, references, prior-work samples

1 owner

One scientist accountable end-to-end

Open

Methods we can describe without a client name

Request NDA and CV

Questions

What scientific leads usually ask first.

What does a first engagement look like?

A 2–6 week fixed-scope sprint with an explicit deliverable (report, pipeline, figures, or diligence memo), success criteria, and a kill switch. NDA first. No long-term retainer required to start. Straightforward analyses can ship in a couple of weeks; multi-modal work takes longer.

Do you handle PHI / HIPAA-regulated data?

Yes, under a signed BAA, typically inside your HIPAA-eligible cloud tenancy. We are HIPAA-aligned and BAA-ready — not HIPAA certified. We do not take custody of regulated data unless the workflow requires it, and we will not accept PHI through the public form or ordinary email.

Who actually does the work?

One senior scientific owner, end to end. Full CV after NDA. Specialists, if needed, are named under that NDA. No bait-and-switch between the first call and delivery.

How do you handle IP and data ownership?

You own deliverables and analyses specific to your data and hypothesis. We retain general-purpose methods and internal tooling that pre-exist or are built independently — standard SOW language, negotiable to pharma and university templates.

Are pipelines reproducible?

Version-controlled, containerized, documented with parameter provenance. Deliverables include the code, the environment, and the report so a reviewer can reproduce the steps.

Can you support academic labs and manuscripts?

Yes. Grant-funded work, publication-grade figures, methods sections, and co-authorship where the contribution merits it are a core part of the practice.

Can you work with GEO, TCGA, MIMIC, NEISS, SEER, GTEx, or CMS?

Yes. Public corpora are a working strength: credentialing, cohort extraction, harmonization (OMOP CDM where relevant), and analysis, following each dataset’s data-use agreement. That is experience with sources — not a claim those projects are clients.

How do you price?

Fixed-scope sprints quoted as a flat fee tied to defined deliverables. Longer work can shift to a monthly retainer. We can work against pharma MSAs and university purchase-order frameworks. We do not publish dollar amounts on this site.

What is Smriti?

Smriti is Biosync’s product: a self-hosted agent memory layer (one binary, offline) where every claim cites a source, facts carry valid-from / valid-until, contradictions surface, and model calls replay bit-identically. A product conversation can run in parallel with a 2–6 week consulting sprint. We do not publish a price; start with a discovery call. Do not put PHI in the public form or the public demo — Smriti is meant to run on your hardware. Product page · Open a live chain.

Do you offer FDA-certified pipelines?

No. We build auditable, version-pinned NGS and document workflows — containerized QC, CI/CD, provenance a reviewer can replay. That is a regulatory-grade posture, not a certification, and we do not name sponsors on this site.

What is outside scope?

Wet-lab work, direct clinical operations, and ownership of regulatory submissions. We work alongside the teams that own those functions — supplying the analysis, the methods, and the defensible evidence — not in place of them.

Next step

Two sentences on the decision you need to defend.

If it is a fit, we send the NDA, then a scoped sprint. Use the form (it opens your mail client) or write hello@bio-sync.tech directly. Do not attach PHI.

NDA, discovery, or project: hello@bio-sync.tech