ABACBS National Seminar Series

The ABACBS national seminar series aims to highlight the work of bioinformaticians across the spectrum of career stages, located in both urban and regional universities.

The seminars are held every few months from 12-1pm EST on Mondays via Zoom. Each seminar features two speakers, with each talk running for approximately 25 minutes, followed by 5 minutes of Q&A time.

The next seminar is scheduled to take place Tuesday October 15th, 12pm-1pm AEST.

Register for the next seminar via https://abacbs.org/seminarzoom

The ABACBS National seminar series is organised by: ABACBS Post-doc Subcommittee.

2024 Seminar program

OCtober 15, 2024

Title: A vision of translational computational pharmacogenomics.

Speaker: A/Prof Michael Menden

Abstract: The talk focuses on the development of biostatistical and machine learning frameworks applied to biomedical data, to retrieve insights in the aetiology of complex diseases and identify patient stratifications. For this, we explore deep molecular characterised biomedical datasets, environmental factors, and tailor our models depending on disease specific knowledge gained through literature and data driven analyses, thus empowering the next generation of precision medicine. In this talk, I will exemplify this with our recently published OncoBird framework (Ohnmacht et al. Nature Communications 2023).

About the speaker: A/Prof Michael Menden is a leader in Computational Pharmacogenomics and commenced an Associate Professor role at the University of Melbourne in August 2023. Previously, he was a Group Leader at Helmholtz Munich (2019-2023) and Senior Scientist at AstraZeneca (2016-2019). He became an ERC Starting Grant Laureate in 2020 (most prestigious Emerging Leader grant in Europe; €1.5M) and Rising Star in Drug Discovery Award by the University of Cardiff in 2022. His PhD was awarded in Computational Biology by the University of Cambridge in 2016, supported by a prestigious EMBL PhD Fellowship (2.7% success rate).

Title: Charting metabolic interactions between gut microbes.

Speaker: Dr. Vanessa R. Marcelino

Abstract: Microorganisms are constantly exchanging nutrients with each other, and the associations between these cross-feeding interactions and human health are still largely unexplored. Part of the problem is that metabolic interactions in complex microbiomes are challenging to detect. In this talk I will discuss how we have combined genome-scale metabolic modelling with restoration ecology theory to identify the metabolic interactions most affected in disease. We applied our framework to over 1600 individuals and found that the microbiome associated with ten chronic diseases are marked by a significantly loss of metabolic interactions. When applied to a Crohn’s disease study, our analyses identified a lack of species with the ability to consume microbially-derived hydrogen sulphide as the main discriminating feature of disease. This conceptual framework paves the way to leverage community ecology for the development of more effective microbiome manipulation strategies.

About the speaker: Dr. Vanessa Marcelino is a group leader at Melbourne Integrative Genomics and the Peter Doherty Institute (University of Melbourne). She has used a range of model systems to investigate the spatial distribution and ecological roles of organisms, from frogs (BS in Brazil) to seaweeds (joint MSc in Germany/France/Belgium), corals (PhD Melbourne U.), and human-associated microbiomes (postdocs at Sydney U. and Hudson Institute). Her team focuses on the ecology of microbial networks, tackling both applied challenges in the development of therapies and microbiome engineering, and fundamental questions on the eco-evolutionary processes underlying biological interactions.

August 5, 2024

Title: Utility of Kolmogorov-Arnold Networks in understanding the genomic architecture of complex disease.

Speaker: A/Prof Phil Melton

Abstract: Traditional machine learning methods are often insufficient in dealing with complex, natural, high dimensional genomic data. This has led to the application of deep learning approaches that extract features through neural network layers. However, classic deep learning methods often act as black-box systems, making meaningful interpretation difficult. To address this, we utilized a recently developed algorithm, Kolmogorov-Arnold Networks (KANs), that allow for mathematical modelling of internal networks, yielding more readily interpretable results from complex data. For proof of concept, we utilised longitudinal early life data along with genome-wide data from the Raine Study and lipidomic data from the Busselton Health Study. In the Raine Study, we predicted body mass index (BMI) from age 5 to age 26 using genome-wide genotype data and early life factors up to age 5, seven genetic risk scores for BMI, and our own genetic association results. Busselton Health Study lipidomic data were used to predict previous incidence preeclampsia from 1893 women using 596 lipid species. Preliminary results show that KAN models outperform other models when groups include genetic data or genetic risk scores, with an area under the curve of greater than 0.9. The most consistent predictors across all time groups were BMI at age 5 and genetic data. In the Busselton Health Study data, we see improved performance in the prediction of preeclampsia when using KANs versus random forest models. These results highlight the potential of KANs as a powerful alternative to classical deep learning models.

About the speaker:  Associate/Professor Phillip Melton is a Principal Research Fellow in the Menzies Institute for Medical Research at the University of Tasmania with expertise in biostatistics and computational biology. He also holds an adjunct appointment at The University of Western Australia (Senior Research Fellow). Prior to relocating to Tasmania, he was Head of Statistical Genetics at Centre for Genetic Origins of Health and Disease (UWA and Curtin) from 2012-2019. He received his PhD in 2008 in anthropological genetics from the University of Kansas and completed his postdoctoral research in statistical genetics at the Texas Biomedical Research Institute (San Antonio, Texas, USA) from 2008-2012. His current research is focused on the identification of genomic markers associated with complex diseases using integrative approaches through machine learning.

Title: Diagnostics for organ transplantation.

Speaker: Harry Robertson

Abstract: Organ transplantation is the sole treatment for end-stage organ failure, but immune-mediated damage threatens transplant longevity. Current monitoring techniques are invasive and imprecise. Leveraging data-driven approaches, we analyzed 150 datasets comprising over 12,000 samples from heart, lung, liver, and kidney transplants to study human pan-organ allograft dysfunction. Our analysis revealed genes consistently associated with dysfunction, including delayed graft function, acute rejection, and fibrosis, across all four organs. We developed a transfer learning framework that improved classification accuracy by integrating data across organs. Validation with a single-center prospective kidney transplant cohort underscored the clinical potential of our approach. This study demonstrates the ability of machine learning models to generalize across organs and provides a transcriptomic resource for developing pan-organ biomarkers of allograft dysfunction.

About the speaker: Harry Robertson is a final-year PhD student at the School of Mathematics, University of Sydney, and the Centre for Transplant and Renal Research. His research focuses on the application of statistical machine learning and deep learning methods to biomedical data analysis, particularly in organ transplantation. During his PhD, Harry has been awarded prestigious scholarships such as the Westpac Future Leaders Scholarship for Technology and Innovation, and a Fulbright Scholarship to continue his research at Harvard Medical School starting August this year.

May 27, 2024

Title: Investigating the use of Protein Structural Information to Understand and Predict Disease

Speaker: Dr Stephanie Portelli

Abstract: The optimization of gene sequencing techniques over the last few decades enables the efficient identification of genetic mutations associated with phenotypes ranging from genetic disease to drug resistant infections. However, this efficiency now presents scientists with a data overload problem, as identified variation cannot be as efficiently confirmed causative through traditional molecular techniques. To address this, we have developed a structural bioinformatics pipeline which initially analyses the effects of confirmed causative missense mutations, compares them against their neutral counterparts, and uses those insights to predict the phenotype of novel, clinically-encountered mutations. By characterising coding mutations within their 3-dimensional protein structure through estimations of stability and affinity changes, we were able to elucidate distinct mechanisms of ALS development across different genes. Similarly, we have highlighted how mutations across different drug targets lead to resistance in tuberculosis, and applied those insights to infer resistance development risk from circulating variation in SARS-CoV-2.   

Using similar principles and insights within a machine learning pipeline, we developed predictive tools for detecting antimicrobial resistance to rifampicin, which has utility across four clinically-indicated infectious diseases. We have also had success in applying our methodology to human genes SMAD4, where we were able to detect cancer-causing variation, and to PTEN, where we successfully distinguished different clinical phenotypes caused by missense mutations. Our approach in mutation analysis has important clinical utility in detecting disease states while also providing insights into the molecular mechanisms of disease in a high-throughput manner, which are invaluable for development of novel therapeutic strategies.

About the speaker:  Dr Stephanie Portelli is a computational biologist specialized in the integration of protein structure and machine learning techniques to develop predictive tools for more informed clinical decision-making. She graduated as a pharmacist at the University of Malta, which she supplemented with a MSc in Drug Design from University College London, where she developed a deeper understanding of the genetic components of antimicrobial drug resistance. She then pursued her PhD at the University of Melbourne, which she finished in 2021, where she studied the effects of missense mutations in both infectious diseases, where they lead to drug resistance and non-infectious diseases, like cancer. She is now a postdoctoral researcher at the University of Queensland where she continues this research. 

Title: The Genome Taxonomy Database in 2024; challenges and opportunities

Speaker: Professor Phil Hugenholtz

Abstract: The Genome Taxonomy Database (GTDB) provides a complete (species to domain) phylogenetically consistent genome-based taxonomy for prokaryotic genomes sourced from the NCBI Assembly database. It is currently in its 8th release (08-RS214) and incorporates ~400K genomes. The next release (09-RS220) scheduled for April 2024 will include ~600K genomes, an increase of 50%. This rapid increase in database size presents a number of challenges, chief amongst them inference of reliable reference trees. After some initial missteps, GTDB has consolidated its approach to nomenclature, and primarily should be viewed as a standardised taxonomic framework. Recently it was suggested that diversity estimates (number of taxa) are inflated in GTDB due to the presence of undetected contamination in genomes. Ranks in GTDB are normalised by relative evolutionary divergence with the exception of species, which is based on an operational ANI threshold of ≥95%. We have been exploring if species can be defined in GTDB by non-operational means that do not use a fixed threshold. GTDB is currently not suited for classification of many pathogens because pathogenic traits often occur at the subspecies level or are horizontally transferred between species, which is not captured in the strictly vertical classification scheme used by GTDB. It is possible that the GTDB workflow can be extended beyond prokaryotes to other life forms such as fungi. In this talk I will present and discuss these challenges and opportunities, and speculate on what the next few years may look like for the GTDB resource.

About the speaker: Beginning with the recognition that we have been ignorant of most microbial diversity due to a strong cultivation bias, I have systematically directed my research to characterise “microbial dark matter” with the ultimate goal of a holistic understanding of microbial evolution and ecology. From 2004 to 2010, I directed the Microbial Ecology and Metagenomics Programs at the DOE Joint Genome Institute (JGI) in the US. In 2010 I returned home to establish the Australian Centre for Ecogenomics, which comprises 50 researchers/core staff who conduct ecogenomics research across a wide range of environmental, engineered and clinical ecosystems underpinned by a genome-based evolutionary framework.