Statistical Genomics Foundry
Paper
domain/comparative-annotationtopic/liftover

Comparative Annotation Toolkit (CAT) — Fiddes et al. 2018

1. Citation

Fiddes IT, Armstrong J, Diekhans M, Nachtweide S, Kronenberg ZN, Underwood JG, Gordon D, Earl D, Keane T, Eichler EE, Haussler D, Stanke M, Paten B. 2018. “Comparative Annotation Toolkit (CAT)—simultaneous clade and personal genome annotation.” Genome Research 28(7):1029–1038. doi:10.1101/gr.233460.117. PMC6028123. OA URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC6028123/ Software repo (from Methods): https://github.com/ComparativeGenomicsToolkit/Comparative-Annotation-Toolkit

2. Access note

Full text read via PMC (open access). License: ”© 2018 Fiddes et al.; Published by Cold Spring Harbor Laboratory Press”; distributed exclusively by CSHL Press for the first six months, then available under Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0). Published 2018 → well past the 6-month embargo, so CC BY-NC applies. Quoting mode: own-words — CC BY-NC resolves to own-words-only in license-policy.yml (NC kept out of casts); §7 paraphrased, functional strings kept verbatim. No paywall boundary.

3. Thesis (1 sentence)

CAT is a fully open-source toolkit that uses a reference-free progressive whole-genome alignment to simultaneously annotate entire clades and identify orthology relationships, including for personal diploid human genomes. (Abstract)

4. Problem & context

Genome annotation traditionally proceeds one genome at a time and is often reference-based. CAT reframes annotation as a comparative problem driven by a Progressive Cactus alignment, which — unlike most earlier alignment methods — is not reference based and includes duplications, making it suitable for many-to-many orthology. Goal: annotate many genomes at once, consistently, and transfer/discover gene models across a clade.

5. Method / approach

Required inputs: a HAL (hierarchical alignment format) multiple-genome alignment produced by Progressive Cactus, plus a GFF3 annotation for the previously annotated genome(s).

Component tools CAT integrates (the load-bearing list):

  • TransMap — projects existing annotations between genomes using the Progressive Cactus alignment; projections filtered by a user-tunable minimum-coverage flag, then the single highest-scoring alignment is chosen.
  • AUGUSTUS, run in up to four parameterizations:
    • AugustusTM / AugustusTMR — the two transMap-projection-based modes. AugustusTMR adds RNA-seq hints (RNA-seq helps fill in information missing from the alignment and resolve evolutionary changes).
    • AugustusCGP — comparative (Comparative Gene Prediction); runs comparative prediction simultaneously across all aligned genomes; ab initio; RNA-seq additionally helps prevent false positives that are inherent to ab initio gene finding.
    • AugustusPB — uses long-read RNA-seq to find novel isoforms (PacBio Iso-Seq). (Paper: CAT runs AUGUSTUS in as many as four distinct parameterizations — two based on transMap projections (AugustusTMR) and two performing ab initio predictions (AugustusCGP and AugustusPB).)
  • homGeneMapping — uses the Cactus alignments to transfer features such as annotations and RNA-seq support between the input genomes.
  • Consensus-finding algorithm (CAT’s own) — merges every source of transcript evidence into a single annotation set, evaluating transMap transcripts against user-tunable RNA-seq / annotation-support flags, then considering ab initio transcripts by locus assignment and their contribution of novel splice junctions supported by RNA-seq or Iso-Seq.

Orchestration: CAT is constructed using the Luigi workflow manager (https://github.com/spotify/luigi), with Toil used for computationally intensive steps best submitted to a compute cluster. (NOT Snakemake — see §12.)

Outputs: diagnostic plots, an annotation set for each target genome, orthology mappings, and a UCSC comparative assembly hub ready to load in the UCSC Genome Browser.

6. Key claims / findings

  • CAT can improve rat annotation, annotate the great apes, annotate a diverse set of mammals, and annotate personal diploid human genomes. (Abstract)
  • Rat: consensus gene set represents an increase of 14,675 genes and 74,308 transcripts over Ensembl V90; 5104 genes and 32,157 transcripts over RefSeq; 14,541 genes and 81,022 transcripts over MAKER2.
  • Great apes: an average of 1677 novel isoforms and 64 novel loci were found across the assemblies with at least one Iso-Seq read supporting the prediction.
  • Progressive Cactus alignments are reference-free and include duplications → suitable for many-to-many orthology (contrast with reference-based aligners).
  • Annotating multiple genomes simultaneously and consistently solves a key scalability issue and yields orthology mappings as a byproduct. (Discussion)

7. Load-bearing statements (own-words — CC BY-NC is own-words-only per license-policy.yml; tool/format names kept verbatim as facts)

  1. (Abstract, thesis) CAT is a fully open-source toolkit that annotates entire clades simultaneously and identifies orthology relationships.
  2. (Results, inputs) the only required inputs are a HAL-format multiple genome alignment (as produced by Progressive Cactus) and a GFF3 annotation for the previously annotated genome(s).
  3. (Results, components) depending on input parameters, CAT runs AUGUSTUS in up to four parameterizations — two relying on transMap projections (AugustusTMR) and two performing ab initio prediction (AugustusCGP, AugustusPB).
  4. (Methods, orchestration) CAT is built on the Luigi workflow manager, with Toil handling compute-intensive steps suited to cluster submission.
  5. (Introduction, advantage) Progressive Cactus alignments are not reference-based and include duplications, making them suitable for annotating many-to-many orthology relationships.

8. Stated scope, assumptions, limitations

  • Requires a Progressive Cactus / HAL alignment and at least one existing GFF3 annotation as substrate — CAT does not build the alignment itself.
  • Novel-isoform / novel-locus discovery depends on RNA-seq and long-read (Iso-Seq) evidence being available; novel calls are supported by ≥1 Iso-Seq read.
  • Ab initio prediction (AugustusCGP) carries false-positive risk that RNA-seq hints mitigate but do not eliminate.

9. Failure modes / invalidity patterns

  • transMap projections are filtered by a user-tunable minimum-coverage flag; low coverage → projection dropped, so poor/incomplete alignments propagate to missing annotations.
  • Consensus inclusion is gated on user-tunable RNA-seq / annotation-support flags; without RNA-seq or Iso-Seq evidence, ab initio novel splice junctions are not admitted.
  • [summarizer-inferred] Because CAT annotates relative to provided reference annotation(s), errors or biases in the input GFF3 can propagate across the clade via projection.

10. What the source does NOT address

  • Does not describe LiftOff (LiftOff post-dates 2018; absent from the paper — see §12).
  • Does not use Snakemake.

11. Open questions / ambiguities

  • Paper collectively parenthesizes the two transMap-based modes as “(AugustusTMR)” while the tooling distinguishes AugustusTM (no RNA-seq) vs AugustusTMR (with RNA-seq); the main-text sentence names only the RNA-seq variant explicitly.

12. Guidance answers

  • What CAT integrates (exact set): TransMap (projection over the Cactus alignment) + AUGUSTUS in up to four modes (AugustusTM / AugustusTMR = transMap-based; AugustusCGP, AugustusPB = ab initio) + homGeneMapping + CAT’s own consensus-finding algorithm. Orchestrated by Luigi + Toil.
  • LiftOff discrepancy (load-bearing): LiftOff is NOT part of 2018 CAT — the word “LiftOff” appears nowhere in the paper. CAT’s projection engine is TransMap, not LiftOff. (LiftOff, Shumate & Salzberg, was published in 2020, post-dating this paper — that dating is general knowledge, not from this source.) Any claim that CAT = “AUGUSTUS + TransMap + LiftOff” is wrong on the LiftOff element.
  • Cactus/HAL substrate: Yes — the only required inputs are a HAL alignment produced by Progressive Cactus plus a GFF3 for the annotated genome(s). Verbatim in §7 #2.
  • Workflow engine: Luigi (Spotify) + Toil — not Snakemake. The guidance’s “Snakemake/Toil-driven” is half wrong: Toil yes, Snakemake no; the top-level manager is Luigi. Verbatim in §7 #4.
  • Outputs / use cases: per-genome annotation sets, orthology mappings, diagnostic plots, UCSC comparative assembly hub; novel isoform/locus discovery. Use cases: rat, great apes, diverse mammals, personal diploid human. (§5, §6)
  • Advantage over single-tool / single-reference projection: the reference-free, duplication-aware Cactus alignment supports many-to-many orthology and simultaneous consistent multi-genome annotation, with a consensus step merging all evidence sources — vs. a single reference-based projection. Verbatim in §7 #5.
  • Repo/config invocation: repo URL captured (§1); no explicit CLI invocation command shown in the main text.