Computational Metabolomics: From Cheminformatics to Machine Learning

Motivation

Mass spectrometry is the predominant analytical technique for detection, identification, and quantification in metabolomics experiments. Technological advances in mass spectrometry and experimental workflows during the last decade enabled novel investigations of biological systems on the metabolite level. Metabolomics started as the study of all metabolites in a living cell or organism; in comparison to transcriptome and proteome, the metabolome is a better proxy of metabolic activity. Emerging fields including personalized medicine and exposomics have expanded the scope of metabolomics to “all” small molecules, including those of non-biological origin. Advances in instrumentation plus rapid increase in popularity, throughput and desired compound coverage has resulted in vast amounts of both raw and processed data; the field is in desperate need for further developments in computational methods. Methods established in other -omics fields are frequently not transferable to metabolomics due to the structural diversity.

Close collaborations between experimental and computational scientists are developing, greatly supported through previous Dagstuhl Seminars on this topic (e.g. Seminars 15492 and 17491). Much progress has been made in the last decade, and the field has started to prosper. However, there are still many problems that we have to overcome. Firstly, there is the issue of parallel and compartmentalized scientific societies: Cross talk between “classical” cheminformatics and novel approaches from bioinformatics and machine learning need to be strengthened. Secondly, hurdles for entering the field are still very high, particularly for scientists from machine learning who do not want to invest time on issues such as data processing. Thirdly, it is difficult to establish and maintain scientific discussions and collaborations between experimentalists, chem/bioinformaticians and machine learners, as they “do not speak the same language”.

This seminar will concentrate on incorporating and bridging the diverse computational communities (experimentalists, cheminformatics, bioinformatics, machine learning), identifying obstacles and developing solutions and new/existing collaborations to overcome them.

Topics to be discussed may include, but are in no way limited to: workflows and strategies for annotation of whole metabolomes; metabolomics and systems biology; mining “expert knowledge”; combining genome and metabolome mining; cheminformatics approaches for partial identification and core structures; searching for novel compounds; retention time information; prediction of ionisation forms and response; separation and identification statistics; the future of untargeted metabolomics.