Karthik Gangadharaiah, independent consultant in Amsterdam. I work directly with research groups, biotech teams and data teams. MSc Molecular Life Sciences, Wageningen University.
Whole genome, long-read and metagenomic projects run end to end: library and run QC, assembly diagnostics, variant calling and GWAS on Slurm or Kubernetes with reproducible workflows.
Recent work includes Oxford Nanopore assembly diagnostics and delivery of high-throughput sequencing pipelines at a commercial genomics facility in Amsterdam.
Computational binder design with RFdiffusion and ProteinMPNN, filtered by molecular dynamics in GROMACS before bench validation. Honest reporting on physical stability and simulation bounds.
Self-hosted workflow engines with n8n and Postgres, FastAPI services, knowledge graphs in Neo4j, and MCP servers that let AI agents query your data securely on your infrastructure.
Open-source computational biology and structural inference repositories with transparent limits.
System Pipeline from de novo backbone generation to all-atom MD validation. 800+ RFdiffusion scaffolds targeting the RBX1 RING domain, ProteinMPNN sequence design, and a 100 ns GROMACS production run (OPLS-AA, TIP3P) on GCP L4 spot instances.
Result Backbone RMSD converged to ~0.2 nm over 100 ns with stable interface hydrogen bonding. HPC/container orchestration included.
Limits In silico validation on a single lead trajectory; binding affinity requires experimental assay confirmation.
System Score-based diffusion with Diffusion Posterior Sampling (DPS) and a differentiable forward model (CTF modulation, Radon projection) across SE(3)-equivariant point clouds and 3D volumetric U-Nets at < -5 dB SNR. Batched FP16 cut per-volume latency up to 4.3×.
Result <0.8 Å aligned RMSD in point-cloud stage; successfully reconstructed unseen folds (hemoglobin alpha, flavodoxin) from synthetic projections.
Limits Tested on synthetic Gaussian noise with known projection angles; sampling requires 50–1,000 steps.
System Neo4j graph unifying ClinVar, Reactome, Open Targets, ChEMBL, GTEx eQTLs (54 tissues), and AlphaGenome variant impact scoring. Deployed with FastAPI and an MCP server (6 tools) for autonomous AI agents.
Result Complete Dockerized deployment with 5-fold cross-validation and topological link prediction (Adamic-Adar, Resource Allocation).
Limits Research prototype. Graph link scores generate prioritized candidate hypotheses, not confirmed drug-target interactions.
KoWerk, Amsterdam
2025 – Present
Oxford Nanopore QC, assembly diagnostics, Slurm and Kubernetes pipeline orchestration, GWAS workflows, and custom research software contracts.
Biomanufacturing Startup
2025 – 2026
Co-developed a nanopore sequencing platform for measuring tRNA charging in biomanufacturing. Program recognized as Nucleate Global Virtual Activator 2026 Top 4 Finalist.
Commercial Genomics Facility, Amsterdam
2022 – 2025
Delivered large-scale metagenomics, single-cell, and repeat GWAS pipelines. Supported ISO 17025 accreditation and client technical scoping.
Wageningen University
2017 – 2021
Thesis in BioNanoTechnology (light-responsive micelles, click synthesis, cryo-TEM) and research in Physical Chemistry & Soft Matter (elastin-like fusion proteins and protein cage nanostructures).
RGUHS
2011 – 2015
Pharmacology, medicinal chemistry, and drug metabolism kinetics.
Describe the data, the bottleneck or the target. I reply by email.