The amount of information that can be collected by a multi-omics analysis of fermentation (e.g. combined metabolomics, transcriptomics, and proteomics experiments) or the information that can be collected from sensors in an industrial bioprocess can easily amount to terabytes of data. Accommodating this massive outpouring of information is one of the defining challenges for the computational analysis of biological systems. The goal of Deep Learning is to effectively analyze these data to make meaningful predictions of the behaviors of complex systems and to identify specific actionable interventions for productive alterations to maximize process outputs. Deep Learning is an Artificial Intelligence approach that accomplishes this by mimicking the activity of layers of neurons in a living brain to solve otherwise intractable computational problems. Applications of Deep Learning to sustainable industrial bioprocesses include analysis of bacterial genomes for the rational genomic engineering for improved efficiency in industrial applications or for the rational design of novel biological processes for new bioproducts. Statistical modeling of biological systems using machine learning and artificial intelligence approaches strongly compliments the results of mechanistic biological models. Deep Learning directly benefits ABF by creating a bridge between the laboratory and computational analysis, streamlining the process by which meaningful information can be extracted from biological observations and rapidly turned to actionable information that can help guide continuous improvements to developing bioprocesses or the real-time optimizations of ongoing industrial fermentations. Deep learning requires substantial amounts of raw data for training and validating AI. This restriction can be partially relieved by considering other model-generated data type (e.g. metabolic models or gene regulation predictions) as input in addition to direct biological observations.
References and Additional Information:
PE Larsen, D Field, JA Gilbert (2012). Predicting bacterial community assemblages using an artificial neural network approach. Nature methods 9 (6), 621.
PE Larsen, Y Dai (2015). Metabolome of human gut microbiome is predictive of host dysbiosis. Gigascience 4 (1), 42
JL Metcalf, ZZ Xu, S Weiss, S Lax, W Van Treuren, ER Hyde, SJ Song (2016). Microbial community assembly and metabolic function during mammalian corpse decomposition. Science 351 (6269), 158-162
Argonne National Laboratory
Lawrence Berkeley National Laboratory