Vivofold Research

Validated Methods & Scientific Software

We implement published bioinformatics and cheminformatics methods from their papers, test them against simulated data with a planted truth, and ship them in Nayancode software with their limits documented. The aim is analysis software that is private, reproducible and honest about what it can do.

A workflow as a directed acyclic graph, in nf-core process naming: every step is versioned, containerised and re-runnable.

Questions

What we want to know

  1. How much of standard genomics and cheminformatics can run privately in a web browser, and at what scale?
  2. How should scientific software prove that it gives correct answers, release after release?
  3. How can every analysis produce its own methods section and provenance automatically?

Approach

How we work on it

  • Clean-room re-implementation from the primary literature, under permissive licences
  • Planted-truth simulations and regression tests on real reference data
  • Web Workers and WebAssembly for local, private computation
  • Automatic provenance capture and methods generation

Methods already implemented

Global and local alignment with affine gapsRead mapping and SNV calling (minimap2 + samtools, WebAssembly)Maximum-likelihood phylogenetics with model selectionSNN-Louvain clustering with clean-room UMAP and t-SNEComBat batch correction, ORA and preranked GSEAInteraction profiling, pockets, SASA and TM-score

Open projects

Projects you can join

01

Browser-scale read mapping benchmarks

Characterise accuracy, memory and runtime of WebAssembly mapping across genome sizes and devices.

02

Numerical regression suites for analysis kernels

Extend planted-truth tests to every statistical kernel and publish the suite.

03

Accessible scientific visualisation

Keyboard and screen-reader access for interactive plots of large datasets.

Collaborate

Propose a project in validated methods & scientific software

Sent securely to [email protected]. Your details are used only to process this request.