Figure 2: Machine learning workflow.
Figure 3: An overview of the structure of a dataset. This
figure demonstrates a labeled dataset because one or more target values
are reported. Datasets are usually prepared by data scientists. However,
the more one knows about a dataset, the easier the machine learning
process will be.
Figure 4: The role of cross-validation in machine learning.First, the generalization power of the ML model is evaluated through
cross-validation. Then, hyperparameter tuning is performed before
training the model to refine the model’s parameters which are called
hyperparameters. Grid search and Bayesian optimization algorithms are
the most common search-based methods to tune hyperparameters. Finally,
the curated model is used to evaluate the prediction capability of
unseen test data.
Figure 5: Integrating constraint-based modeling and machine
learning. CBM and ML can be integrated in different ways for analysis
and optimization of fermentation parameters. (a) Predicting
parameters by ML when fluxomics are used as inputs. (b)Predicting parameters by ML when the integration of fluxomics with
multi-omics is used as input. (c) Predicting fluxomics and
constraint-based models by ML when multi-omics are used as inputs.