RNA-velocity discourse 2025–2026 (tweet digest)

A digest of six bookmarked tweets tracking the field conversation — new methods, the skeptics, a benchmark, and the RegVelo announcement. Each is a short pointer; depth is in the linked papers (not yet ingested). Useful for situating FlowVelo in the current discourse and for the velocity-skepticism thread.

Summary

The 2025–2026 conversation has two poles. New methods keep arrivingCell2fate (biophysical module decomposition), spVelo (multi-batch spatial velocity), and RegVelo (GRN-coupled) — while prominent skeptics sharpen the critique: LiorPachter calls velocity-on-UMAP “biological gobbledygook”, and GennadyGorin’s biophysics PhD reframes velocity around normalization and the chemical master equation. A UCSF benchmark stresses that method choice and data quality dominate outcomes. The split mirrors the xing-hu-regvelo-debate: build more methods vs fix the theoretical foundations.

The six bookmarks

  • Cell2fate — @naturemethods, 2025-03-04 (Bayraktar & Stegle labs, Nat Methods 2025): “improves RNA velocity analysis … by module decomposition of realistic biophysical models of transcription dynamics.” A biophysics-forward method. → s41592-025-02608-3.
  • LiorPachter critique — @lpachter, 2025-08-08: “Another paper showing that putting arrows on UMAPs (RNA velocity) is biological gobbledygook. How many more such papers are needed?” Amplifies velocity-benchmark-17studies (now ingested) — whose own abstract is far more measured than his gloss. The loudest external skeptic.
  • spVelo — @LukasValihrach, 2025-08-18 (Genome Biology 2025): “RNA velocity inference for multi-batch spatial transcriptomic data.” A spatial-velocity sibling to TopoVelo. → s13059-025-03701-8.
  • GennadyGorin — @GorinGennady, 2026-03-03: PhD centerpiece in Nat MethodsMonod (monod-gorin, now ingested): CME-based stochastic modeling of nascent+mature counts that minimizes distortive normalization. Pachter-lab biophysics-of-velocity line.
  • RegVelo announcement — @fabian_theis, 2026-05-12: “unify RNA velocity + GRNs into one model → better OOD prediction of perturbations (gene KOs), incl. neural-crest KO predictions.” Author’s framing of RegVelo. (cf. the xing-hu-regvelo-debate critique.)
  • UCSF benchmark — @RNASeqBlog, 2026-06-04: “Comparing RNA velocity methods for tracking cell development … method selection and data quality matter.” Now identified + ingested: Ancheta et al., PLoS Comput Biol 2026 (velocity-benchmark-ancheta-2026), CZ Biohub SF / UCSF — 5 methods × 3 datasets, no winner, hypothesis-generating. A second reliability benchmark alongside velocity-benchmark-17studies. → 10.1371/journal.pcbi.1014303.

Physical-time grounding (standing lens)

A discourse digest, not a method, but it maps where the physical-time pressure is coming from:

Connections

Contradictions

  • No factual contradiction. The skeptic bookmarks reinforce physical-time-grounding; the new-method bookmarks confirm the field is advancing on non-temporal axes.

NOTE: 進度(2026-06-12 更新)——六則 bookmark 全部 full-ingest 完成:Pachter 放大的 benchmark (= velocity-benchmark-17studies,full PDF)、Gorin 的 Monod(= monod-gorin,full text)、 Cell2fate、spVelo、RegVelo announcement(= regvelo)皆已處理;UCSF benchmark 已認出並 ingest = velocity-benchmark-ancheta-2026(Ancheta et al., PLoS Comput Biol 2026)。「omission of cell growth」critique(= cell-growth-omission-2025)仍是 abstract-level(bioRxiv API only)。本 digest 的 full-text backlog 已清空。