CTO & Co-Founder, Skolyn AB · Postdoctoral Research Fellow, Uppsala University · Seconded National Expert, European Research Council Executive Agency
Clinical Bioinformatician, Linköping University Hospital · Data Science Specialist, DTU Bioengineering
Location: Greater Linköping Metropolitan Area, Sweden (also Baku, Azerbaijan)

Links: LinkedIn · ORCID · Google Scholar · ResearchGate
Contact: olafylimanov@skolyn.se · olaf.imanov@it.uu.se · +46 76 236 80 88

• Overview
• Now / Roadmap
• Appointments
• Research and Engineering Themes
• Selected Publications
• Selected Artifacts
• Technical Stack
• Talks, Teaching, Service
• Citation and Archiving
• Disclaimer

I build clinically deployable machine learning systems at the intersection of medical imaging, clinical genomics, and multi-omics.
My work spans research and engineering: problem formulation → reproducible pipelines → interpretable modeling → multi-site validation → deployment in regulated settings.

Core priorities:

• Interpretability and auditability as first-class requirements (not post-hoc add-ons)
• Privacy-preserving evaluation across institutions under real heterogeneity
• Reproducible computational biology pipelines aligned with clinical operations
• Interoperability boundaries (DICOM/NIfTI, HL7/FHIR) and traceable system behavior

• Clinical decision support: strengthen traceability, structured explanations, and audit logs suitable for clinical stakeholders.
• Federated benchmarking: finalize multi-site evaluation protocols, site-effect analysis, and reporting conventions.
• Clinical genomics: harden Nextflow/nf-core execution patterns (provenance, version pinning, reproducible environments) for routine diagnostics.
• Publications: complete and submit pending manuscripts in medical imaging, sonography interpretability, and multi-omics biomarker discovery.

• Deployment engineering: improve model monitoring, dataset shift detection, and calibration strategies in real operational constraints.
• Open-source artifacts: extract and publish reusable components (evaluation harnesses, reference implementations, reproducibility scaffolding).
• Human-AI collaboration: formalize reader-study and interaction evaluation packages for clinical workflows.

• Multi-modal and multi-omics integration at scale with clinically meaningful interpretability (concept-level and pathway-level reporting).
• Federated learning beyond training: federated evaluation and governance patterns with auditable compliance-friendly reporting.

• Skolyn AB (CTO & Co-Founder) — AI-driven clinical decision support; interoperable deployment (HL7/FHIR); scalable inference (Python/PyTorch; Docker/Kubernetes)
• European Research Council Executive Agency (Seconded National Expert) — evaluation workflows and strategic analysis for large-scale research portfolios
• Uppsala University (Postdoctoral Research Fellow) — medical imaging, neuroradiology, explainability; federated learning benchmarking across Swedish hospitals
• Linköping University Hospital (Clinical Bioinformatician) — clinical genomics operations; rare disease diagnosis; nf-core/Nextflow workflows
• DTU Bioengineering (Data Science Specialist) — proteomics and systems biology; graph learning for protein interactions

Previously: Google Health (Research Scientist; Technical Program Manager II) · Finnish Center for Artificial Intelligence (Senior Research Scientist)

• Interpretability: SHAP, Integrated Gradients, attention-based analyses; structured explanation reporting
• Reliability: calibration, uncertainty estimation, OOD behavior, subgroup and shift analysis
• Human-AI collaboration: evaluation protocols, reader studies, interaction design constraints

• Robust training under institutional heterogeneity (site effects, device/protocol variation)
• Privacy-preserving evaluation design and reporting (reproducible metrics across sites)
• Operational constraints: governance, logging, monitoring, update strategies

• Rare disease workflows: variant calling/annotation/interpretation and reporting practices
• Multi-omics integration: network analysis, graph neural networks, latent factor models
• Reproducibility: Nextflow/nf-core conventions; provenance, parameterization, environment pinning

• Imaging: DICOM, NIfTI; reproducible pre-processing and annotation practices
• Health data: HL7/FHIR integration patterns; system boundaries and audit trails

1. Imanov, O.Y.L., Schmidt, M., Andersson, L. (2025). Integrative machine learning framework for early-stage neurodegeneration biomarkers from multi-omics profiles. Journal of Proteome Research (forthcoming).
2. Imanov, O.Y.L., Chen, E., Kumar, R. (2025). Explainable AI in sonography: transferring interpretability from proteomics to medical imaging. JAMIA (forthcoming).
3. Imanov, O.Y.L., Christensen, A.N., Nielsen, M. (2025). Designing adaptive human-AI systems for collaborative problem solving in fetal ultrasound diagnostics. Medical Image Analysis, 85:102745.
4. Imanov, O.Y.L., Kaski, S., Virtanen, P. (2024). Adaptive federated ensembles for heterogeneous medical imaging datasets. IEEE Transactions on Medical Imaging, 43(12):4234–4247.
5. Chen, E., Imanov, O.Y.L., Zhang, W. (2024). Human-in-the-loop evaluation of large language models for medical question answering. Nature Digital Medicine, 7(1):245.
6. Imanov, O.Y.L., Bergström, S., Andersson, J. (2023). Trustworthy reinforcement learning with human-in-the-loop feedback for medical imaging. Medical Image Analysis, 89:102876.

Full list: https://scholar.google.com/citations?hl=en&user=WSqps1YAAAAJ

The sections below follow a consistent structure: Scope, Non-goals, Reproducibility, and Citation.
For clinical or medically adjacent code, explicit boundaries are stated to reduce misuse risk and to clarify intended usage.

Repository: https://github.com/skolyn/skolyn-ai

Scope

• Reference implementations and platform foundations for explainable clinical decision support workflows.
• System engineering patterns for low-latency inference, traceable outputs, and integration boundaries.

Non-goals

• Not a substitute for professional medical judgment.
• Not a standalone EHR system and not a comprehensive clinical workflow product by itself.
• Not a claim of regulatory clearance; regulatory readiness work is handled as a separate governance track.

Reproducibility

• Prefer containerized execution for deterministic environments.
• Where applicable: explicit data splits, pinned dependencies, and evaluation scripts producing stable metrics.

Citation

• Add a CITATION.cff file at repository root (recommended).
• If archival is enabled, attach a DOI to releases and reference it in the repository documentation.

Repository: https://github.com/fcai/fedmed · PyPI: https://pypi.org/project/fedmed/

Scope

• Federated learning baselines and evaluation harnesses for heterogeneous clinical datasets.
• Utility modules for reporting, aggregation experiments, and reproducible comparisons.

Non-goals

• Not a production-grade FL orchestrator; focuses on research-grade baselines and benchmarking.
• Not a privacy guarantee by default; privacy claims must be explicit, threat-modelled, and tested.

Reproducibility

• Run scripts should produce (a) metrics tables, (b) calibration and error analyses, and (c) a manifest of configs.
• Prefer: fixed seeds, version pinning, and explicit dataset provenance statements.

Citation

• Provide CITATION.cff and versioned release tags; archive key releases to obtain a DOI when appropriate.

Repository: https://github.com/nf-core/clinicalgenomics

Scope

• Community pipeline ecosystem for clinical genomics workflows (Nextflow).
• Contributions typically focus on: robustness, portability, and reproducible execution in clinical operations.

Non-goals

• Not a single-institution bespoke pipeline; aims to be generalizable and configurable.
• Not a substitute for local clinical validation requirements.

Reproducibility

• Follow nf-core guidelines: parameterized execution, container profiles, and consistent provenance outputs.

Citation

• Use the project’s recommended citation entry and add local documentation for clinical validation context.

Languages: Python, R, SQL
ML: PyTorch, TensorFlow; CNNs, Transformers, GNNs; evaluation and calibration
MLOps/Infra: Docker, Kubernetes; containerized training/inference; reproducible environments
Pipelines: Nextflow, nf-core; CI-ready workflow patterns
Standards: HL7/FHIR, DICOM, NIfTI

Talks (2023–2025): ECR · MICCAI · ICML · NeurIPS Workshops · Nordic AI in Medicine Summit · Google Health AI Symposium
Teaching (2025, Linköping University): data analysis with Python; regression/statistical modeling; probability and statistics; multivariate methods
Review/service: MICCAI, NeurIPS, ICML, ICLR; Nature Digital Medicine, Medical Image Analysis, IEEE TMI, Bioinformatics, PLOS Computational Biology

Recommended for research software:

• Add a CITATION.cff file to repositories to provide machine-readable citation metadata.
• For long-lived releases, archive versions and issue DOIs (e.g., via Zenodo GitHub integration).
• Document how to reproduce key results (configs, seeds, environment, and evaluation entrypoints).

This profile reflects personal work and research. It is not an official product statement.
Last updated: December 2025