Ever since Robert Hooke and Antoni van Leeuwenhoek discovered the existence of microscopic organisms, our understanding of how they affect our health has advanced tremendously. From finding out about their life cycles and their significance within the various environmental niches they colonize, to utilizing them in various industrial processes, our ability to harness microbes in immunization and treatment of diseases makes the concept of microbiology quite interesting.
Microbiology is the study of micro-organisms like archaea, bacteria, fungi, protozoa and viruses. Going past the superficial knowledge of these micro-organisms, microbiology takes a deeper exploration into understanding their biochemistry, ecology, evolution, cell biology, clinical significance as well as their host agents. It is super amazing how organisms as small as microbes can play important roles in every aspect of our lives.
Starting out, microbiologists viewed microorganisms with microscopes and grew them in a culture which serves as a suitable medium for their growth but overtime research scientists discovered that this method was costly and labor intensive and overtime scientific inventions led to sequencing technologies. Due to the invention of this high thorough-put sequencing technology, a large amount of microbial data is being generated thus, machine learning method has steadily won its way into the area of microbial analysis as the go-to method to solve classification and interaction problems.
Machine Learning Methods in Microbial Analysis
Machine learning is the process of teaching a computer system how to efficiently make accurate predictions when given data and these methods are divided into supervised and unsupervised learning. In supervised learning, the model is trained using a training set while unsupervised learning draw inferences from data with insignificant human supervision without pre-existing labels.
To develop a machine learning algorithm, four main steps are involved which are: extraction of feature, obtaining operational taxonomic units table (OTU) via clustering, selection of important features to increase accuracy and efficiency and finally training the model using a training data set. An operational taxonomic unit or OTU is used to classify groups of closely knitted micro-organisms and it is an important step in the analysis of microbial data especially when analyzing small subunits of 16s and 18s ribosomal RNA data set. The OTU table which contains: the types of operational taxonomic units as well as individual species annotation information and the quantities for each samples “ can be obtained after OTU clustering and species classification.
Feature Dimensionality in Microbial Analysis
It is important to remember that feature dimensionality reduction is an important part of this data analysis because some microbes have higher data dimensions that others thus, some common techniques to reduce this dimensionality are principal components analysis (PCA) and principal co-ordinates analysis (PCoA). PCA reduces the data dimensionality while retaining most of the data variation while PCoA selects the most significant co-ordinates.In microbial studies, supervised learning methods such as k nearest neighbor (KNN), naives bayes (NB), random forest (RF) and support vector machines (SVM) are used.
Classification and Prediction of Microbial Species
The first step to classifying and predicting microbial species accurately is to properly identify microorganisms because various microbes have individual characteristics unique to them. Examples of novel approaches to taxonomic classification that is built upon machine learning principles are IDTAXA, MARVEL, VirFinder and VirSorter. IDTAXA uses both IdTaxa and LearnTaxa functions. LearnTaxa uses tree descent “ a ML method similar to the decision tree to classify each training sequence into its tagged taxon while IdTaxa uses objects returned by LearnTaxa as input data. MARVEL uses the random forest method to predict double stranded DNA bacteriophage. VirFinder uses k-mer method to identify overlapping groups of virus and VirSorter provides a near perfect identification of viral sequences living outside the genetic material of the host. Although all these methods have similar specificity performances, MARVEL has a better sensitivity performance.
Disease Prediction Using Microorganisms
Research has shown that by studying microbial populations, their genes and environment, we can predict the phenotypes of the host and the environment because microorganisms exchange information with the surrounding environment and host cells. With this information, we can better protect the host. Present in the human body are many microbial communities and once a particular community of microbes is out of balance or suffers foreign body invasion, the human body is likely to fall ill. By analyzing microbial communities, we can better understand the disease and then make effective decisions concerning treatment. For example, Beck and Foster (2014) used the genetic algorithm (GP), random forest (RF), and logistic regression (LR) to predict Bacterial Vaginosis — a disease associated with the vaginal microbiome by identifying its microbial communities.
Although many puzzles about the application of machine learning in microbiology still remain unresolved, and will require collaboration with researchers within the health, sciences and informatics niche, microbial studies has travelled beyond just learning about microbes only to the understanding and improvement human health.
