{
  "id": 14810225,
  "title": "A Benchmark of Unsupervised Circadian Phase Inference Methods for Bulk Transcriptomic Data",
  "url": "https://urgent.news/2026/10/15/a-benchmark-of-unsupervised-circadian-phase-inference-methods-for",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-15T00:00:00.000Z",
  "source": {
    "name": "bioRxiv",
    "slug": "biorxiv",
    "url": "https://www.biorxiv.org/content/10.64898/2026.10.08.757747v1?rss=1"
  },
  "original_language": "en",
  "account": "A comprehensive evaluation of ten unsupervised circadian phase inference methods and two clock-gene baselines on 25 bulk transcriptomic datasets with reference times has been conducted. The primary metric used was the aligned circular mean absolute error (MAE), with additional analyses including permutation tests, circular correlations, and gene-level reconstruction of rhythmicity, acrophase and amplitude. CHIRAL demonstrated the lowest mean MAE of 1.80 hours, with a 95% confidence interval of 1.21-2.97 hours. The two clock-gene baselines had MAEs of 2.15 and 2.20 hours, and the mean errors of the next four methods (TEMPO, CYCLOPS, and CYCLOPS 2.0) fell between 2.49-2.80 hours. The remaining six methods had mean errors of 3.91-4.86 hours, with confidence intervals of these six methods substantially overlapping. The accuracy of the methods varied depending on the dataset, clock-gene rhythm strength, random seed, and adjustment for batch or covariates. When the inferred phases were applied to analyze gene rhythms in 22 datasets with reference times, the four most accurate methods identified rhythmic genes with a median ROC AUC of 0.64-0.70 and accurately recovered gene acrophases with median errors of 0.18-0.64 hours. However, the agreement between the inferred phases and external reference acrophases varied between methods and tissues, particularly in two mouse datasets without collection times.",
  "summary": "Circadian phase can be inferred from gene expression when sample collection times are unavailable, but the available methods have not been systematically compared. We compared ten unsupervised phase inference methods and two clock-gene baselines on 25 datasets with reference times (16 real datasets from five studies and 9 simulations). The primary measure was the aligned circular mean absolute…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}