Classification

Machine learning using Galaxy webinar

22 - 26 June, 2020

​

Institute of Applied Biosciences (INAB)

Centre for Research and Technology Hellas (CERTH)

Thessaloniki, Greece

2

The ML taxonomy

3

Supervised Learning

4

Supervised Learning

• Apply what has been learned in the past to new data using labeled examples to predict future events. �
• Starting from the analysis of a known training dataset, the learning algorithm produces a prediction model that can provide targets for any new input (after sufficient training). �
• The learning algorithm can also compare its output with the correct, intended output and find errors in order to modify and improve the prediction model accordingly.

5

Classification

Two steps:

• Building the classifier or model:
• the learning step, in which the classification algorithms build the classifier.
• built from the training set made up of database samples and their associated class labels.
• Each sample that constitutes the training set is referred to as a class.�
• Applying the classifier to a classification task:
• Classifier is used for classification
• Test data is used to estimate the accuracy of classification rules

6

Classification vs Regression

https://aldro61.github.io/microbiome-summer-school-2017/sections/basics/

7

Classification vs regression

8

Linear regression

• Regression algorithms can be used for example when some continuous value needs to be computed as compared to classification where the output is categorical.
• So whenever there is a need to predict some future value of a process which is currently running, regression algorithm can be used.
• Operating on a two dimensional set of observations (two continuous variables), simple linear regression attempts to fit, as best as possible, a line through the data points.
• The regression line (our model) becomes a tool that can help uncover underlying trends in our dataset.
• The regression line, when properly fitted, can serve as a predictive model for new events.
• Linear Regressions are however unstable in case features are redundant, i.e. if there is multicollinearity
• Example where linear regression can be used are:
• Using gene expression data to classify (or predict) tumor types using gene expression data

9

Decision Trees (Supervised)

• Single trees are used very rarely, but in composition with many others they build very efficient algorithms such as Random Forest or Gradient Tree Boosting.
• Decision trees easily handle feature interactions and they are non-parametric, so there is no need to worry about outliers or whether the data is linearly separable.
• Disadvantages are:
• Often the tree needs to be rebuilt when new examples come on.
• Decision trees easily overfit, but ensemble methods like random forests (or boosted trees) take care of this problem.
• They can also take a lot of memory (the more features you have, the deeper and larger your decision tree is likely to be)
• Trees are excellent tools for helping to choose between several courses of action.
• Example: Classification of genomic islands using decision trees and ensemble algorithms

10

Random Forest (Supervised)

• Random Forest is an ensemble of decision trees.
• It can solve both regression and classification problems with large data sets.
• It also helps identify most significant variables from thousands of input variables.
• Random Forest is highly scalable to any number of dimensions and has generally quite acceptable performances.
• However with Random Forest, learning may be slow (depending on the parameterization) and it is not possible to iteratively improve the generated models
• Random Forest can be used in real-world applications such as:
• Predict patients for high risks for certain diseases

11

Support Vector Machines (Supervised)

• Support Vector Machine (SVM) is a supervised machine learning technique that is widely used in pattern recognition and classification problems — when your data has exactly two classes.
• Advantages include high accuracy and even if the data is not linearly separable in the base feature space, SVM can work well with an appropriate kernel.
• However SVMs are memory-intensive, hard to interpret, and difficult to tune.
• SVM is especially popular in text classification problems where very high-dimensional spaces are the norm.
• SVM can be used in real-world bioinformatics applications such as:
• detecting persons with common diseases such as diabetes
• Classification of genomic islands

12

Validation of supervised ML algorithms results

• To test the performance of the learning system:
• The system can be tested with sequences where the labels are known (and were excluded from the training set because they were intended to be used for this purpose).
• Based on the results of the test data, the performance of the learning system can be assessed.

13

Some terms

• Confusion matrix��
• Precision��
• Specificity��
• Recall / Sensitivity��
• Receiver Operating Characteristic (ROC) and AUC curves

14

https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html

1 - Specificity

Sensitivity