Skip to content

ROADMAP_TRAJ.md

Latest commit

 

History

History
94 lines (73 loc) · 4.51 KB

ROADMAP_TRAJ.md

File metadata and controls

94 lines (73 loc) · 4.51 KB
Raw

Trajectory-Inference R-Package Roadmap

Ranked candidates for the next series of py-<pkg> ports under github.com/omicverse. Ordered by citation density (citations/year) × benchmark importance, excluding anything already published under omicverse/.

Status legend

  • ✅ done (under omicverse/)
  • 🟡 in progress
  • ⬜ not started
  • ⛔ skip (already has a Python implementation OR is integrated in scanpy)

Already-done at omicverse (do NOT re-port)

Package Repo Algorithm class
Monocle 2 py-monocle2 ordinal (pseudotime)
Slingshot omicverse.single._pyslingshot ordinal
Palantir omicverse.externel.palantir ordinal
VIA omicverse.via ordinal
Diffusion map (homebrew) omicverse.single._diffusionmap ordinal
CytoTRACE 2 omicverse.single._cytotrace2 ordinal
CellFateGenie omicverse.single._cellfategenie inference
STT omicverse.externel.STT ordinal
scdiffusion omicverse.externel.scdiffusion ordinal

Tier 1 — Highest priority (must-port for any TI benchmark)

Rank R package Year Citations cites/yr Algorithm class Status Seed template
1 tradeSeq 2020 ~911 ~152 inference ⬜ py-miloR
2 URD 2018 ~982 ~123 ordinal ⬜ py-monocle2
3 TSCAN 2016 ~547 ~55 ordinal ⬜ py-monocle2
4 destiny (DPT) 2016 ~341 ~34 ordinal ⬜ py-monocle2

Notes on Tier 1:

  • tradeSeq is the highest-ROI port — DE-along-trajectory is the standard companion to every TI method. Pairs with Slingshot which is already in omicverse.
  • URD has the highest citation count (Science 2018) but most are biological applications; method-method citations are ~30% of total.
  • TSCAN is the simplest port (PCA + Mclust + MST + principal curve) — best for verifying the engineering loop.
  • destiny is the canonical DPT reference; scanpy's DPT is an independent reimplementation, not a port — a real py-destiny would be valuable.

Tier 2 — Strongly recommended

Rank R package Year Citations cites/yr Algorithm class Status Seed template
5 SCORPIUS 2016 ~165 ~17 ordinal ⬜ py-monocle2
6 condiments 2024 ~38 ~19 inference ⬜ py-miloR

Notes:

  • SCORPIUS is the dynbenchmark linear-trajectory winner; same authors as Slingshot/tradeSeq team.
  • condiments is the 2024 Nat Commun multi-condition trajectory framework — "occupy" while citations are still building.

Tier 3 — Optional (specialised topologies / specific use cases)

Rank R package Year Citations cites/yr Algorithm class Status
7 SLICER 2016 ~184 ~18 ordinal ⬜
8 psupertime 2022 ~44 ~11 ordinal ⬜
9 Totem 2023 ~17 ~6 ordinal ⬜
10 Mpath 2016 ~91 ~9 ordinal ⬜
11 PhenoPath — — — ordinal ⬜
12 ouija — — — ordinal ⬜
13 CellTrails — — — ordinal ⬜
14 reCAT — — — ordinal (cell-cycle) ⬜
15 FateID — — — ordinal ⬜
16 RaceID / StemID — — — clustering + ordinal ⬜

Already-Python (do NOT re-port)

  • ⛔ PAGA — native to scanpy
  • ⛔ scVelo — Python
  • ⛔ CellRank — Python
  • ⛔ scFates — Python
  • ⛔ Wishbone — Python
  • ⛔ Wanderlust — Python
  • ⛔ elPiGraph — has a Python implementation (ElPiGraph.P)

Suggested execution order

For a Trajectory benchmark that complements omicverse's existing TI methods:

1. py-TSCAN        (Equivalence test the loop end-to-end on a simple port)
2. py-tradeSeq     (Highest-ROI; needed by every benchmark)
3. py-destiny      (DPT canonical reference)
4. py-URD          (Branching tree topology)
5. py-SCORPIUS     (Linear SOTA; dynbenchmark winner)
6. py-condiments   (Multi-condition extensions)

Each port should follow PROTOCOL.md verbatim and tick through CHECKLIST.md. Update this file's status column after each release.

Source

Citation counts: Google Scholar (May 2026). TSCAN cite from Semantic Scholar. Selection criteria: dynbenchmark top performers + post-2019 high-impact additions + multi-condition / supervised extensions. See the conversation transcript in omicverse-rebuildr/ for the audit trail.