AI Readiness

How the platform makes research data consumable by AI/ML pipelines, with live adoption statistics and scoring details.

Libre Biotech is designed so your research data is structured for machine learning from day one. This page documents the AI-readiness features, how they are scored, and how each platform component contributes to both FAIR compliance and ML consumability.

Component overview

Each platform feature serves both FAIR data principles and AI/ML readiness. The cards below show live adoption statistics from the platform.

AI-Ready Score

Each investigation displays an AI-Ready Score card (0-100) alongside the FAIR Score in the investigation sidebar. The score measures how consumable the data is for ML pipelines across 8 independently scored dimensions.

Scoring dimensions

#	Dimension	Max	What it checks	Source tables
1	Machine-Readable Formats	12	Has studies (6 pts) and processes (6 pts) — the minimum data needed for ISA-JSON or ML-Ready export	studies, process_studies
2	Structured Metadata	12	Has ontology annotations on samples (6 pts) and study factors defined (6 pts)	annotations, study_factors
3	Data Provenance	12	Sample lineage depth (4-8 pts: depth 1-2 = 4 pts, depth 3+ = 8 pts) plus process count (2-4 pts)	process_input_samples (recursive CTE on output_sample_id, depth cap 10), process_studies
4	Standard File Formats	12	Files with standard extensions: FASTQ, BAM, SAM, CRAM, CSV, TSV, GFF3, GFF, GTF, VCF, BED, BigWig, FASTA, JSON, XML (1-4 files = 8 pts, 5+ = 12 pts)	process_files, files
5	Computational Reproducibility	13	Succeeded CWL analysis runs (1-2 runs = 8 pts, 3+ = 13 pts)	analysis_runs WHERE status='succeeded'
6	Quantitative Measurements	13	Measurements with Unit Ontology terms (1-4 = 8 pts, 5+ = 13 pts)	measurements JOIN assays WHERE unit_id IS NOT NULL
7	Consistent Sample Labeling	12	>80% labels match a common prefix pattern (3-6 pts) and >80% have descriptions (3-6 pts)	samples — regex heuristic on labels
8	API Accessibility	13	Investigation visibility: public = 13 pts, group = 8 pts, private = 3 pts	investigations.visibility

Total maximum: 99 points (normalised to 0-100 scale).

Badge thresholds

Score	Badge	Colour
90-100	Excellent	Green
75-89	Good	Blue
50-74	Fair	Yellow
0-49	Needs Work	Red

Suggestions

Each missing criterion generates an actionable suggestion displayed beneath the score card. For example:

• Metadata Add ontology annotations to samples for structured metadata
• Provenance Add sample lineage (process_input_samples links) for provenance tracking
• Reproducibility Run CWL analysis pipelines for computational reproducibility
• Accessibility Set visibility to "public" for unrestricted programmatic access

How to improve your score

To improve	Action
Machine-Readable Formats	Create at least one study and link processes to it
Structured Metadata	Use the annotation panel on sample pages to add ontology terms (organism, tissue, disease). Define study factors (genotype, treatment) on the study page
Data Provenance	Link samples as source materials using the "Link Sources" button on process pages. Build chains: tissue → extract → library → sequencing
Standard File Formats	Attach FASTQ, BAM, CSV, GFF3, or VCF files to processes via the file upload panel
Computational Reproducibility	Submit CWL analysis runs from the Compute → Analyses page and wait for them to succeed
Quantitative Measurements	Record measurements with assays, selecting unit terms from the Unit Ontology (UO) — e.g. nanogram, microliter, RQN
Consistent Sample Labeling	Use a consistent naming pattern (e.g. MOUSE-BRAIN-001, MOUSE-BRAIN-002) and add descriptions to all samples
API Accessibility	Set the investigation visibility to "public" on the investigation edit page

ML-Ready Data Export

The ML-Ready export flattens all investigation data into a single samples × features matrix. This is the primary format for loading platform data into ML pipelines.

How it works

1. Collects all samples across the investigation (via studies → processes → samples)
2. Gathers ontology annotations for each sample, grouped by slot (organism, tissue, etc.)
3. Gathers study factor values for each sample (genotype, treatment, etc.)
4. Gathers quantitative measurements for each sample (RQN, concentration, etc.)
5. Builds a provenance summary per sample using a recursive CTE on process_input_samples (depth capped at 10)
6. Discovers all unique column names from the union of annotation slots, factor names, and measurement types
7. Builds a flat rows matrix where each row is a sample and each column is a feature (nulls for missing values)

Access methods

Method	How
Web UI	Click the ML-Ready button on any investigation page. Use the dropdown to choose CSV, JSON, or ZIP format
REST API (JSON)	GET /api.php/v1/investigations/{id}/ml-export
REST API (CSV)	GET /api.php/v1/investigations/{id}/ml-export?format=csv
REST API (ZIP)	GET /api.php/v1/investigations/{id}/ml-export?format=zip — CSV data + README in a self-documenting ZIP bundle

JSON schema

{
  "metadata": {
    "investigation_id": 3,
    "investigation_title": "Mouse Transcriptomics",
    "export_date": "2026-03-14",
    "sample_count": 42,
    "feature_count": 8,
    "ontology_prefix_map": {
      "NCBITaxon": "http://purl.obolibrary.org/obo/NCBITaxon_",
      "UBERON": "http://purl.obolibrary.org/obo/UBERON_"
    }
  },
  "columns": [
    {"name": "sample_id", "type": "identifier"},
    {"name": "sample_label", "type": "string"},
    {"name": "annotation:organism", "type": "ontology", "ontology_curie": "NCBITaxon:10090"},
    {"name": "annotation:anatomy", "type": "ontology", "ontology_curie": "UBERON:0000955"},
    {"name": "factor:genotype", "type": "categorical"},
    {"name": "measurement:RQN", "type": "numeric", "unit": "UO:0000186"}
  ],
  "rows": [
    [1, "SAMPLE-001", "Mus musculus", "brain", "wild-type", 8.7],
    [2, "SAMPLE-002", "Mus musculus", "liver", "knockout", 7.2]
  ],
  "provenance_summary": {
    "1": {"depth": 3, "process_chain": ["extraction", "sample_prep", "sequencing"]},
    "2": {"depth": 2, "process_chain": ["extraction", "sequencing"]}
  }
}

Column types

Type	Description	Prefix	Example
identifier	Unique integer ID	—	sample_id
string	Free text	—	sample_label
ontology	Ontology term value with CURIE reference	annotation:	annotation:organism
categorical	Discrete category from study factors	factor:	factor:genotype
numeric	Numeric measurement with optional unit CURIE	measurement:	measurement:RQN

CSV format

The CSV export uses the column names from the JSON columns array as the header row. Null values are empty strings. Standard RFC 4180 CSV encoding.

sample_id,sample_label,annotation:organism,annotation:anatomy,factor:genotype,measurement:RQN
1,SAMPLE-001,Mus musculus,brain,wild-type,8.7
2,SAMPLE-002,Mus musculus,liver,knockout,7.2

Code examples

Python: load into pandas

import requests
import pandas as pd

API = "https://librebiotech.org/api.php/v1"
headers = {"X-API-Key": "YOUR_KEY"}

# Option 1: JSON
resp = requests.get(f"{API}/investigations/3/ml-export", headers=headers)
data = resp.json()
df = pd.DataFrame(data["rows"], columns=[c["name"] for c in data["columns"]])

# Option 2: CSV (simpler)
from io import StringIO
resp = requests.get(f"{API}/investigations/3/ml-export?format=csv", headers=headers)
df = pd.read_csv(StringIO(resp.text))

# Access provenance
provenance = data["provenance_summary"]
print(f"Sample 1 lineage depth: {provenance['1']['depth']}")
print(f"Process chain: {provenance['1']['process_chain']}")

R: load into tibble

library(httr)
library(jsonlite)
library(tibble)

api <- "https://librebiotech.org/api.php/v1"
key <- "YOUR_KEY"

resp <- GET(paste0(api, "/investigations/3/ml-export"),
            add_headers("X-API-Key" = key))
data <- fromJSON(content(resp, "text", encoding = "UTF-8"))

df <- as_tibble(data$rows)
colnames(df) <- sapply(data$columns, function(c) c$name)
print(df)

Python: scikit-learn pipeline

import requests
import pandas as pd
from sklearn.preprocessing import LabelEncoder, StandardScaler

API = "https://librebiotech.org/api.php/v1"
headers = {"X-API-Key": "YOUR_KEY"}

resp = requests.get(f"{API}/investigations/3/ml-export", headers=headers)
data = resp.json()
df = pd.DataFrame(data["rows"], columns=[c["name"] for c in data["columns"]])

# Separate features by type
numeric_cols = [c["name"] for c in data["columns"] if c["type"] == "numeric"]
categorical_cols = [c["name"] for c in data["columns"] if c["type"] == "categorical"]
ontology_cols = [c["name"] for c in data["columns"] if c["type"] == "ontology"]

# Encode categoricals
for col in categorical_cols + ontology_cols:
    if col in df.columns:
        df[col] = LabelEncoder().fit_transform(df[col].fillna("unknown"))

# Scale numerics
if numeric_cols:
    df[numeric_cols] = StandardScaler().fit_transform(df[numeric_cols].fillna(0))

FAIR Score comparison

The AI-Ready Score complements the existing FAIR Score. Both are shown in the investigation sidebar:

Aspect	FAIR Score	AI-Ready Score
Focus	Data findability, accessibility, interoperability, reusability	ML/AI consumability of the data
Dimensions	4 (Findable, Accessible, Interoperable, Reusable)	8 (formats, metadata, provenance, files, reproducibility, measurements, labeling, accessibility)
Scale	0-100 (average of 4 sub-scores)	0-100 (sum of 8 dimension scores)
Checks	Metadata completeness, DOI, contacts, licenses, protocols	Ontology annotations, lineage depth, file formats, CWL runs, unit measurements, label consistency
Audience	Data managers, repository curators, compliance officers	ML engineers, AI researchers, computational biologists

For AI/ML Researchers

If you are an AI/ML researcher looking to consume data from this platform, visit the dedicated For AI/ML Researchers page for API quickstart guides, endpoint documentation, code examples in Python/R/curl, and details on data formats and ontology coverage.