skills.rest
K-Dense-AI
K-Dense-AIOFFICIAL
Explore all skills by this author →
Website𝕏@k_dense_aiGitHub

waypoint-bio

Embed, fine-tune, and benchmark microbiome taxonomic profiles with Outpost Bio's Waypoint foundation models.

41.1k|3.8k|Updated Oct 19, 2025
Search Tags:#embeddings#fine-tuning#bioinformatics#taxonomy#metagenomics#microbiome#foundation model
One-click install
npx skills add https://github.com/K-Dense-AI/scientific-agent-skills --skill waypoint-bio
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: waypoint-bio
Source: https://github.com/K-Dense-AI/scientific-agent-skills/tree/main/skills/waypoint-bio
Command: npx skills add https://github.com/K-Dense-AI/scientific-agent-skills --skill waypoint-bio

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires waypoint-bio, pandas, torch, transformers, datasets, peft, scikit-learn, and includes scripts (resource) and references (resource) components.

What problem does it solve?

Applying transformer-based foundation models to microbiome data requires converting profiler outputs (MetaPhlAn, Kraken2, QIIME 2) into a specific format, checking vocabulary coverage, and correctly running embedding, fine-tuning, and benchmarking workflows — with several silent failure modes like dropped out-of-vocabulary taxa and leaky random splits.

Core Features & Use Cases

  • Data Conversion: Convert MetaPhlAn, Kraken2/Bracken, QIIME 2, and MGnify abundance tables into waypoint format with correct lineage prefixes and separators.
  • Embedding & Fine-tuning: Generate fixed-size sample embeddings from Waypoint checkpoints (6m/45m/170m) and fine-tune classification or regression heads with LoRA support and covariate handling.
  • Benchmarking & Pretraining: Score models on the eight-task Compass benchmark or pretrain GPT-2-style models on the 539k-sample Atlas corpus.
  • Use Case: A researcher with stool metagenomic profiles converts Kraken2 reports to waypoint format, checks vocabulary coverage, then fine-tunes Waypoint-45m to predict drug degradation rates with donor-aware data splits.

Quick Start

Use the waypoint-bio skill to convert my MetaPhlAn abundance table into waypoint format and embed the samples with the Waypoint-6m checkpoint.

Frequently Asked Questions about waypoint-bio

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I convert MetaPhlAn or Kraken2 output for Waypoint models?▼

Use the bundled profiler_to_waypoint.py script with --format metaphlan or --format kraken. It rewrites pipe separators to semicolons, maps rank codes to tokenizer prefixes, and normalizes abundances into the Taxa and Relative Abundances list-columns Waypoint expects.

How do I embed microbiome samples with a Waypoint checkpoint?▼

Run waypoint embed with --model set to a Hub id like outpost-bio/Waypoint-6m and --data pointing to a waypoint-format parquet file. Output is one fixed-size vector per sample (256, 512, or 768 dimensions depending on checkpoint) using last-token pooling by default.

Why do I get a 401 or 403 error when loading Waypoint models?▼

The Waypoint checkpoints, Atlas, and Compass are gated Hugging Face repos. You must request access on each repo page individually and then authenticate with hf auth login or an HF_TOKEN environment variable; a token alone without per-repo access approval is not enough.

When should I not use Waypoint for microbiome prediction?▼

Below roughly 1,000 labeled samples, Waypoint underperforms a random forest on raw relative abundances, with the crossover near 10,000 training examples. Fit the random forest baseline first and only adopt the transformer if it demonstrably wins on your data.

Why are some of my taxa silently dropped during embedding?▼

Waypoint's vocabulary is frozen at pretraining time from the Atlas corpus, and out-of-vocabulary taxa resolve to <unk> and are dropped without warning. Run scripts/vocab_coverage.py to measure per-sample taxon and abundance coverage before trusting downstream results.

Does Waypoint support species-level taxonomic resolution?▼

No, the published checkpoints use genus-level tokenization by default, collapsing species-level distinctions. Changing the taxon_rank setting requires re-pretraining a model from scratch, not just re-tokenizing the data.

Related Skills

Explore More Education & Research Skills→
stellaromicsstellaromics

bio-workflows-metagenomics-pipeline

Automate metagenomic analysis from FASTQ reads to taxonomic and functional profiles.

Official
Advanced
nahisahonahisaho
3

scientific-biobank-cohort

Process phenotype dictionaries and GWAS summary statistics for PheWAS analyses.

Community
Intermediate
matheus-rechmatheus-rech
1

forest-plot-creation

Generate forest plots from yi and sei using metafor package in R.

Community
Intermediate
zongtingweizongtingwei
25

microbiome-amplicon

Automate amplicon microbiome workflows from raw reads to abundance summaries.

Community
Intermediate