Skip to main content
TRANSLATIONAL DECISION SCIENCE

Evidence architecturefor biomarker-driven programs

We make measurement validity, inferential design, and computation explicit, so program decisions are defensible to regulators and stakeholders.

Get in touch
How we work

Validity, design, and computation: each constrains what the others can deliver.

Assay precision means little without inferential architecture to support it. Sophisticated pipelines can't rescue underpowered designs. We work at the intersection, where these disciplines constrain and enable each other.

Measurement Validity

What the assay actually quantifies: precision, specificity, and the assumptions underneath.

Experimental Design

Study architecture, power, decision gates, and the inferential chain from observation to claim.

Computational Rigor

Multi-omics integration, reproducible pipelines, and analytical choices that survive scrutiny.

We evaluate where your study is most constrained, whether that's assay behavior, design architecture, or analytical pipeline, and focus effort where it changes outcomes.

The goal isn't methodological purity; it's decisions you can defend.

Services

Measurement, design, and computation applied across the development arc - with stakeholder communication that bridges biology, statistics, and regulatory teams.

EmbeddedIntegrated into your team for ongoing programs
  • Translational Strategy

    Deep biological synthesis across preclinical models - animal model selection, species relevance, mechanistic plausibility. Translating findings into clinical decision frameworks with power analysis and go/no-go clarity.

    Biology SynthesisAnimal ModelsPower Analysis
  • Clinical Design

    Estimand-first trial planning. Treatment effect definition, intercurrent event handling, sample size justification, adaptive approaches, and analysis plans aligned to regulatory questions.

    EstimandsPower AnalysisICH E9(R1)
  • AI/ML & Computation

    Model development with credibility documentation. Validation strategies, performance monitoring, and transparency requirements aligned to FDA AI/ML framework.

    Model CredibilityFDA AI GuidanceReproducibility
  • Biomarker Evidence

    Fit-for-purpose evidence generation. Context of use definition, qualification strategy, natural history studies, and the evidentiary path from marker to endpoint.

    Fit-for-PurposeRWEQualification
Discuss Your Situation

The Biostochastics Method

Systems-level thinking first: pathway integration, network context, multi-scale mechanism. Then systematic evaluation across measurement, design, and computation.

Measurement

  1. Fit-for-Purpose Assessment

    Context of use drives evidence requirements. What decision does this biomarker support, and what analytical validation does that demand?

  2. Qualification Strategy

    Regulatory precedent, evidentiary standards, and the path from exploratory marker to qualified endpoint. Plan the evidence package early.

Design

  1. Estimand Specification

    Define the treatment effect precisely: population, endpoint, intercurrent events, summary measure. ICH E9(R1) alignment from protocol draft.

  2. Decision Framework

    Go/no-go criteria, adaptive decision rules, and the regulatory question the trial must answer. Explicit thresholds force honest assessment.

Computation

  1. Model Credibility

    FDA AI/ML framework alignment: validation strategy, performance monitoring, transparency documentation. Credibility is earned, not asserted.

  2. Reproducibility Architecture

    Version-controlled pipelines, audit-ready outputs, containerized environments. When regulators ask "show your work," you can.

Every engagement produces stakeholder-ready deliverables - not just recommendations, but documented artifacts that biology, statistics, and regulatory teams can act on.

Industry Experience

A track record of translational research across pharmaceuticals, consumer health, and precision medicine.

Work as Founder and Principal Scientist of Biostochastics

Bryleos

Leading a translational research program at the intersection of neurodegenerative, metabolic, and regenerative medicine as well as precision nutrition

Cogni

Market research and translational strategy consulting for wearable-based dementia monitoring

Research Scientist II (Statistical Genetics)

Arivale

Statistical Geneticist developing direct-to-consumer genetic insights and data clouds for personalized biomarker-driven interventions. After Arivale's closure, joined Institute for Systems Biology as Senior Research Scientist supporting industry-facing observational, clinical trials, and in silico work.

Prior Collaborations

At the Institute for Systems Biology, Yale, Baylor, and University of Houston

Olink Proteomics

P3 flow cell protocol development for Olink Explore (with ISB Molecular Core); sample and dataset curation for Olink Insights ranges data

Roche Genentech

Proteomic signatures of treatment response to ocrelizumab in MS

Gilead

Multi-omic and immune effects of remdesivir in acute COVID-19

Thorne HealthTech

Metabolomics-focused machine learning-based assay exploration

Sanitarium

Multi-omics of weight loss resistance

iCarbonX US

Dataset curation, onboarding, multi-omic effects of SSRIs

Post-Finasteride Syndrome Foundation

Multi-omic effects of finasteride on plasma biomarkers

MindTrust

Development and psychometric validation of computerized cognitive assessments in Arabic

Harris County Juvenile Probation

Clinical trial with neurophysiological and epigenetic biomarkers for intensive remediation program with field deployment

Background

Sergey KornilovSergey Kornilov is a computational and translational scientist. Training in quantitative psychology (psychometrics, experimental design, measurement theory) shapes how Biostochastics approaches study design, biomarker evidence, and computational validation. “What are we actually measuring?” comes before any analysis.

Fifteen years across academia, research non-profits, and industry. A path from cognitive neuroscience through neurophysiology to molecular genetics and multi-omics, from behavioral phenotypes to molecular mechanisms. That breadth enables the systems-level integration Biostochastics brings: pathway context, biological mechanism, and regulatory alignment considered together. Current focus on translating multi-omic signatures into therapeutic insights for neurodegenerative and metabolic disease.

At the Institute for Systems Biology, led industry collaborations from study design through computational analysis for clinical trials and observational studies. Multi-omic characterization of pharmacodynamic effects across medications, nutraceuticals, and behavioral interventions, contributing to discovery of a novel biomarker of statin efficacy and characterizing effects of anti-CD20 monoclonal antibodies. At Arivale, built genetic insights and data infrastructure for a personalized intervention program. Currently leading translational R&D at Bryleos.

How I work: Every engagement produces documented artifacts: decision rationale, evidence architecture, analytical choices and their justification.

Track Record
  • Currently leading translational R&D at Bryleos
  • Fifteen years of research and R&D across academia, research non-profits, and industry
  • Dual PhDs in Experimental and Educational Psychology (UConn, Moscow State University)
  • Post-doctoral training at Yale University School of Medicine, Baylor College of Medicine
  • Former Research Assistant Professor, University of Houston; Research Scientist, Saint-Petersburg State University
  • Led industry collaborations at the Institute for Systems Biology for Genentech, Gilead, and others
  • Statistical Geneticist at Arivale, building direct-to-consumer genetic insights
  • 65+ publications, 5k+ citations, including Cell, Nature, Nature Biotechnology
  • Over $1.5M in industry research contracts; ~$5M total in grant funding
  • Established the role of SETBP1 in common neurodevelopmental disorders of language in a genetically isolated population, validated in multiple independent cohorts using behavioral and neuroimaging data (SRCD Outstanding Doctoral Dissertation Award)
  • Current Editor at Academia Neuroscience and Brain Research
Google Scholar
Website
ORCID

Capabilities

Infrastructure for reproducible computation, methods for rigorous analysis, and fluency in current regulatory frameworks.

Compute & Pipelines

Cloud Infrastructure

GCP · AWS · HPC

Containerization

Docker · Kubernetes

Workflow Orchestration

Snakemake · Nextflow

Reproducibility

targets · Quarto · Shiny

Analysis & Methods

Languages

Python · R

Single-Cell & Spatial

Seurat · Scanpy · CellRanger

Machine Learning

scikit-learn · XGBoost · PyTorch

Trial Statistics

Estimands · Adaptive designs · Bayesian · Survival

Multi-Omics Integration

Custom pipelines · mixOmics

Regulatory Alignment

Quality Systems

GxP-aware · 21 CFR Part 11

Clinical Standards

ICH E9(R1) · CDISC · SDTM

FDA AI/ML Framework

Credibility documentation · Validation

FAQ

Good faith engagement with hard questions is the minimum.

Measurement validity, experimental design, and computational analysis, integrated. Deep phenotyping across modalities: digital phenotyping, neuroimaging, and molecular phenotyping (proteomics, metabolomics, transcriptomics). Systems biology: pathway integration, network context, multi-omics synthesis. Fluent in regulatory frameworks: ICH E9(R1) estimands, FDA AI/ML credibility guidance, fit-for-purpose biomarker qualification. Domain focus on metabolic and neurodegenerative disease. I don't run clinical operations. I build the quantitative and biological architecture underneath.

You shouldn't outsource your entire program to me, and I don't offer that. I work on evidence architecture: measurement, inference, and validation questions where methodological rigor matters. Quantitative neuroscience training instills hypervigilance about replication, effect sizes, and measurement validity—rigor often under-emphasized in translational biology. The replication crisis hit behavioral sciences first; those who survived learned to see how studies fail. That vigilance, applied to biomarker validation and study design, catches things domain experts miss. The biology is your domain.

Core: neurodegenerative disease, neurodevelopmental disorders, cognitive neuroscience. Neuroimaging including EEG and fMRI. Multi-omics: transcriptomics, proteomics, metabolomics. Technical: statistical genetics, biostatistics, machine learning. Applied: target identification, measurement models, systems biology, precision medicine. Actively developing: metabolic disease, target validation. Depth varies—deepest in neurological disease, neuroimaging, and multi-omic integration. I'll tell you when we're at my edges.

Based in Seattle, working remotely with teams worldwide. Async-first with scheduled syncs as needed. Timezone flexibility for US, Europe, and Asia-Pacific. Zoom, Slack, and Discord native; tolerant of Microsoft Teams and Google Meet. Project-based delivery with concrete artifacts: the study design memo, the biomarker strategy, the statistical framework. On-site visits possible for longer partnerships.

Expert networks are transactional: $1,000 to $1,200/hour for one-hour calls, often without continuity, program context, or artifacts. Judgment without accountability, answers without deliverables. This is project-based delivery: context, artifacts, follow-through. The value is in the work product, not the call.

Limited by design. One to two concurrent projects maximum. High-stakes protocol design requires deep work. The commitment is to fewer engagements at higher depth, not maximizing billable hours. Lead times vary, so ask. If you need 40 hours/week, this probably isn't the right fit. 24/7 availability can be arranged under the right conditions.

AI is indispensable. Representative tools: Elicit and Consensus for literature synthesis, PandaOmics and Cheiron for target discovery, knowledge graphs for pathway analysis, plus custom pipelines for project-specific needs. What AI cannot replace: architectural judgment under uncertainty, cross-domain integration where constraints conflict, and accountability for reproducible artifacts. AI provides options; humans take responsibility.

You'll get clear documentation of why and what the options are. "Not fixable" is a valid finding, sometimes the most valuable one. Some programs should be stopped, not continued. Honesty about what's possible is the service. The value is in knowing which is which.

Programs where methodological rigor is the bottleneck—where the biology might be right but the evidence architecture is fragile. Leadership willing to engage with uncomfortable findings. Not for programs seeking validation of decisions already made.

Contact

Let's discuss how I can help with your biostatistics, multi-omics, or translational research challenges.

sergey@biostochastics.com

What to expect:

  1. You describe your situation (program stage, need, timeline)
  2. 30-minute conversation to assess fit - no charge
  3. If mutual fit: scoping and terms
  4. Work begins, typically within 1-3 weeks

Not every inquiry becomes an engagement. Fit matters on both sides.