AIBN Virtual Seminar Series: Application of mixture models to RNA-seq data to discover ageing regulators

We are pleased to host AIBN PhD Student, Sasdekumar (Sas) Loganathan from the Mar Group to talk about his PhD work as part of the AIBN Virtual Seminar Series. Please take this opportunity and join us virtually.

Date: Thursday, 22 October

Time: 12pm - 12:40pm

Venue: Online Via Zoom

The seminar is free, but registration is essential. To register please click here.

Abstract

Ageing is a complex process where the combined effects of environmental and genetic factors make it challenging to isolate specific regulators of ageing. Moreover, both the variation in ageing between individuals and the variation between different tissues make the task of identifying regulators even more of a challenge. Previous studies that have modelled gene expression have been successful in identifying regulatory information by detecting novel genes and pathways. Mixture models can model variability through the detection of multimodality at the gene level from RNA-sequencing (RNA-seq) data. Therefore, we used mixture models to interrogate this variability to study the regulatory programs involved in ageing. 

We applied mixture models using three different distribution-based assumption to transcriptome-wide RNA-seq data for four human tissues (subcutaneous adipose, skeletal muscle, skin and whole blood) from the Genotype-Tissue Expression (GTEx) cohort. We identified lists of candidate genes that clustered according to multimodal distributions with donors that showed significant changes in age. A large percentage of these genes have also been detected by standard differential gene expression (DE). However, genes that are unique to mixture models (MMUG) (ie. genes that are only detected by mixture models), have a similar percentage overlap to known known ageing databases as genes discovered by DE. Furthermore, pathway over representation analysis of these MMUG on Hallmark, Reactome and Gene ontology (GO) databases have detected pathways not detected by DE genes such as, TNFA signalling via NFKB in skeletal muscle. Many pathways were common between MMUG and DE genes, indicating that mixture models potentially can detect different parts of the pathway such as TNFA signalling via NFKB. 

The results indicate that modelling gene expression variability using mixture models can help uncover regulators that have a potential role in understanding human ageing. The application of unsupervised clustering from mixture models identified genes that had significantly altered ages between donors that corresponded to different modes. Showcasing the importance of not assuming a single mode when analysing RNA-seq data to model a gene’s expression distribution in a cohort. Using resources like the Digital Ageing Atlas and GenAge we confirmed that some of these genes have been previously implicated in ageing. Most genes not in these resources have also been implicated in ageing. Lastly, the results also indicate that mixture models detecting different genes in a pathway, which might lead to potential drug targets.

Bio

Sasdekumar (Sas) graduated with a degree in biomedical science in 2011 from Murdoch university. After completing the degree he worked as a research assistant for in a drug discovery laboratory where he developed multiple imaging assays for the various life cycle stage of the Malaria parasite. These assays were used to test multiple small molecule libraries and lead optimisation of potential hits.

He started his PhD in 2019 under the supervision of Associate Professor Jessica Mar, where he has been using statistical models to analyse RNA-seq to better understand ageing.

Don't wait to register. Register today to avoid missing your chance to tune in!