Supervision team information: Computational Biology & Gene Regulation group, Norwegian Centre for Molecular Biosciences and Medicine (NCMBM)
Supervisor: Anthony Mathelier; e-mail address: anthony.mathelier@ncmbm.uio.no
IFI supervisor: Pooya Zakeri; Geir Kjetil Ferkingstad Sandve
Internship place: NCMBM, Oslo, Norway
Keywords: dimensionality reduction, matrix factorization, transcription factor, computational biology
Project description
Transcription factors (TFs) control gene expression through sequence‑specific binding at TF binding sites (TFBSs). Accurately inferring which TFs are active—and which subsets of their TFBSs are driving that activity—remains a central challenge. Existing approaches often summarise activity at the gene level or integrate across multiple omics layers, which can obscure TF‑specific signals at the binding‑site level. Inspired by gene pathway‑level latent variable modeling such as MAYA [1] and PLIER [2], this Master’s project will implement a multi‑modal (multi‑mode) decomposition to the TF level: given one omics dataset (e.g., CAGE, ATAC‑seq, or a histone‑mark profile), the method will infer TF activity and the active subsets (“modes”) of TFBSs that best explain the observed signal. Specifically, the candidate will implement and assess dimensionality reduction methods such as principal component analysis, non-negative matrix factorization, sparse principal component analysis, and other matrix factorization methods, to identify regulatory modules based on TF activities.
This TFBS‑centric framing offers two advantages. First, it preserves positional specificity, allowing us to ask which sites of a TF are active in a given condition. Second, it yields interpretable latent “modes” (e.g., promoter‑proximal vs distal, strong‑motif vs weak‑motif, cofactor‑co‑bound subsets), providing mechanistic hypotheses rather than a single undifferentiated TF score.
Leveraging our expertise in computational biology and machine learning, and our databases, notably JASPAR [3] and UniBind [4], the candidate will build the computational software to decompose TF activity from omics data. This approach will be applied to in-house datasets to investigate regulatory dynamics underlying breast cancer and cellular reprogramming, providing insights into transcriptional regulation with potential clinical and biological implications.
Learning outcomes
The Master student will gain experience in regulatory genomics, matrix factorization, semi‑supervised learning with biological priors, robust evaluation, and reproducible software engineering.
References
[1] Multi-modal quantification of pathway activity with MAYA. Y. Landais and C. Vallot. Nature Communications (2023)
[2] Pathway-level information extractor (PLIER) for gene expression data. Mao et al. Nature Methods (2019)
[3] JASPAR 2024: 20th anniversary of the open-access database of transcription factor binding profiles. I. Rauluseviciute, R.R. Puig, et al. Nucleic Acids Research (2024)
[4] UniBind: maps of high-confidence direct TF-DNA interactions across nine species. R.R. Puig, P. Boddie, A. Khan, J.A.C. Mondragon, A. Mathelier. BMC Genomics (2021)