Participation of RSAT team to the IBIS challenge (2024)

Evaluation of RSAT motif discovery tools for the IBIS challenge 2024.

https://ibis.autosome.org/

Support of the challenge presentation teleconference (video + slides): https://disk.yandex.ru/d/82FEnwPn158pog

Software environment installation

Instructions to install the required software environment can be found in INSTALL.md.

Submission for the final board

Wrap-up report

Documents

Doc	Path
Method wrap-up	reports/IBIS-challenge-final-stage_method-write-up_RSAT-team_2024-09-01.docx
Final PWM	submissions/final/PWM.txt
Zip archive	submissions/final/RSAT-team_IBIS-2024_final_submission.zip
URL to the zip archive with the matrix file (PWM.txt )	https://github.com/jvanheld/IBIS_2024/raw/main/submissions/final/RSAT-team_IBIS-2024_final_submission.zip
md5sum of this zip archive	f37d12e2af6b4a3e7247afd4801b3ac3
URL of the method write-up	https://github.com/jvanheld/IBIS_2024/raw/main/reports/IBIS-challenge-final-stage_method-write-up_RSAT-team_2024-09-01.docx

How to reproduce our results?

All results are generated using make scripts in the makefilesdirectory.

Detailed instructions to launch these make scripts are provided in the HOWTO.md file.

Availability

This material is available

• on github: https://github.com/jvanheld/IBIS_2024/
• tagged versions are archived on zenodo [ADD LINK]

Methods

Motif discovery approaches

We ran different approaches and statistical models to detect exceptional motifs.

• oligo-analysis: over-represented k-mers

• position-analysis: k-mers with positional bias, i.e. their distribution along the sequences differs from a homogeneous distribution (chi-square test)

• dyad-analysis: over-represented dyads, i.e. pairs of 3-mers with a spacing of a given width (all widths from 0 to 20 were considered)

These 3 methods return k-mers or dyads declared significant according to these criteria. These k-mers then serve as seeds to generate position-specific scoring matrices (PSSM) based on a scanning of the train sequences.

Motif clustering

The different motif discovery methods return similar yet slightly different motifs, which are partly redundant. We use the RSAT program matrix-clustering to cluster the motifs at different stages :

• Dataset-wise clustering. Motifs discovered by peak-motifs for a given dataset. An interesing criterion is to check whether motifs discovered by the 3 different algorithms are clustered into a single cluster, which indicates that these motifs are both over-represented (oligo-anlaysis and dyad-analysis) and positionally biased relative to the peak centers (a priori, we expect to see a concentration of the TFBS close to the peak centers).

• TF-wise clustering. We cluster all the motifs discoverd for a given transcription factor in the different data types (CHS, GHTS, HTS, SMS, PBM). An interesting criterion is to check if a motif is robust to the experimental method used to characterize binding sites.

Analysis of motif enrichment in train sequences

• matrix-quality

Motif selection criteria

The selection of motifs to be submitted for the IBIS challenge relied on multiple criteria

• detected by several motif discovery algorithms. For CHS and GHTS datasets, we priorize motifs showing both over-representation and positional bias (oligo-analysisand position-analysis). The other approaches return sequences too short for position-analysis to be relevant.

• peak coverage : percent of the peaks that contain at least one instance (site) of the motif (PSSM)

• For motifs declared significant by position-analysis, visual inspection of the positional distribution of occurrences to check that they are enriched near the peak centers (in principle , the algorihtm can also return centrally empoverished motifs, or any othe rtyep of positional heterogeneity of the occurrences)

• enriched in the train sequences versus test sequences (matrix-quality)

• we discard poor complexity motifs

• ...

Motif optimisation

[TO BE WRITTEN]

Motif selection

[TO BE WRITTEN]