GénOncoPGxBio

Bioinformatics & AI

Bioinformatics & AI in genomics

Watch scope

Tools, algorithms and AI models applied to constitutional genetics and genomic diagnostics. Weekly selection on pathogenicity predictors (AlphaMissense, PheMART, EVE, SpliceAI...), long-read sequencing (PacBio HiFi, Oxford Nanopore), large language models (LLMs) applied to clinical genomics, structural variant and CNV callers, diagnostic pipeline benchmarks, and computational phenotyping methods (HPO, PhenoAI).

Inclusion / exclusion criteria

Included
  • ✓Tools with clinical validation or benchmark published on real-world data
  • ✓AI models applied to patient cohorts with rare diseases
  • ✓Pipelines usable in genomic diagnostic context
  • ✓Computational phenotyping methods
  • ✓New sequencing approaches improving diagnostic yield
Excluded
  • ✕Purely theoretical methodology without application to clinical genomic data
  • ✕Algorithmic optimization without demonstrable clinical relevance
  • ✕Tools limited to basic research
  • ✕Publications on entirely synthetic data

Scoring methodology

Each article is evaluated out of 10 points. Inclusion threshold: score ≥ 5.

  • Clinical impact (0–3)
    Primary criterion. +3: changes practice immediately (CPIC A dose recommendation, prophylactic surgery, deployable tool). +2: likely impact in the short term. +1: indirect contribution to counselling or understanding. 0: fundamental interest only.
  • Strength of evidence (0–3)
    +3: robust causality — functional + multi-family segregation, large prospective cohort, or RCT. +2: good evidence — quality functional alone, cohort ≥10 families or ≥100 patients. +1: partial evidence. 0: preliminary data or single case report.
  • Novelty (0–2)
    +2: unprecedented element (new gene, new CPIC A/B interaction, breakthrough tool). +1: significant extension (phenotypic expansion, VUS→P/LP reclassification, major replication). 0: minor replication or narrative review.
  • Cohort size / robustness (0–1)
    +1 if ≥5 independent cases, ≥100 genotyped patients, cohort ≥1000, open-source code (bioinformatics), or meta-analysis ≥5 studies. 0 otherwise.
  • Journal quality (0–1)
    +1 for a peer-reviewed journal recognised in the specialty (Nat Genet, AJHG, Genet Med, JCO, Pharmacogenomics J, Nat Methods…). 0 otherwise.
  • Preprint penalty (−1)
    −1 for bioRxiv/medRxiv preprints not yet peer-reviewed. The preprint is still included if it carries major information.
Domain-specific note: In bioinformatics & AI, clinical impact is the primary criterion (0–3 pts): a tool deployable in routine diagnostics with demonstrated yield improvement on real patient data scores +3. Strength of evidence (0–3 pts) distinguishes rigorous benchmarking on real clinical data from synthetic-only validation. Open-source code gives the robustness bonus (+1).

Sources consulted

  • —PubMed + bioRxiv — bioinformatics, genomic AI, long-read sequencing query
  • —Genome Research, Genome Biology, Nature Methods
  • —Bioinformatics (OUP), Briefings in Bioinformatics
  • —Nucleic Acids Research (special tool issues)
  • —bioRxiv — bioinformatics preprints (preprint status explicitly mentioned)

Author

Dr. Thibaut Benquey — Medical biologist specialised in constitutional genomics (WGS/WES), leading SASU Geno'X. Generalist approach to constitutional molecular diagnostics, covering the full spectrum of clinical indications.

LinkedInContactKo-fi

Informational document — does not constitute medical advice.

→ Full methodology — constitutional genetics watch