Introduction to bioinformatics
Hoorcollege 6 – 24 januari 2024
Machine learning
What is machine learning?
- Machine learning is a form of Artificial Intelligence
- Finding interesting patterns in “big data” using statistics.
o Machine learning is the study of computer algorithms that improve
automatically through experience. [...] Machine learning algorithms build a
model based on sample data, known as "training data", in order to make
predictions or decisions without being explicitly programmed to do so.
- Examples:
o Email filtering
o Speech recognition: (Alexa/Siri ect.)
o Those personalized add for sneaker that keep following you because you
bought a pair online once.
Example of machine learning: personalized medicine
- What makes patients similar to each other?
- Which patient should get which drug?
- Based on (molecular) profiling of patient
Unsupervised clustering reveals structure in the samples
Supervised learning reveals that good and poor prognosis patient have different gene
expression profiles
We can use this information to predict the outcome of new patients
Good prognosis no need for chemo
Train to predict a model
Classification rule
, Unsupervised learning: discover interesting structure in the
data
- Clustering
- Dimensionality reduction: PCA, tSNE, UMAP
- Uses “unlabeled” data
o One have genes
o Label: yes of no …
o Unsupervised no label
Two groups of patient, one low expressed and one high
expressed
Supervised learning: Make predictions on new data, given
training data:
- Classification & Regression
- Used “labeled” data (i.e. samples from healthy &
diseased patients)
Unsupervised machine learning: clustering
How?
- Hierarchical clustering
- K-means clustering
- Fuzzy clustering
Why?
- Are there subgroups of a disease? (e.g. multiple types of breast cancer, which require
different treatment)
Can we split the data meaningfully into different groups?
Do we find a pattern here?
Example data 2D, normally D is much larger (e.g. number of genes measured), but I can’t
draw that
There is no one correct answer. It’s not clear how many groups are here
Algorithm to put data into two groups
K-means clustering algorithm
Decide how many clusters you should have
This is not always straightforward
1. The algorithm starts by randomly assigning each of your observations to only one of
the k clusters.
Hoorcollege 6 – 24 januari 2024
Machine learning
What is machine learning?
- Machine learning is a form of Artificial Intelligence
- Finding interesting patterns in “big data” using statistics.
o Machine learning is the study of computer algorithms that improve
automatically through experience. [...] Machine learning algorithms build a
model based on sample data, known as "training data", in order to make
predictions or decisions without being explicitly programmed to do so.
- Examples:
o Email filtering
o Speech recognition: (Alexa/Siri ect.)
o Those personalized add for sneaker that keep following you because you
bought a pair online once.
Example of machine learning: personalized medicine
- What makes patients similar to each other?
- Which patient should get which drug?
- Based on (molecular) profiling of patient
Unsupervised clustering reveals structure in the samples
Supervised learning reveals that good and poor prognosis patient have different gene
expression profiles
We can use this information to predict the outcome of new patients
Good prognosis no need for chemo
Train to predict a model
Classification rule
, Unsupervised learning: discover interesting structure in the
data
- Clustering
- Dimensionality reduction: PCA, tSNE, UMAP
- Uses “unlabeled” data
o One have genes
o Label: yes of no …
o Unsupervised no label
Two groups of patient, one low expressed and one high
expressed
Supervised learning: Make predictions on new data, given
training data:
- Classification & Regression
- Used “labeled” data (i.e. samples from healthy &
diseased patients)
Unsupervised machine learning: clustering
How?
- Hierarchical clustering
- K-means clustering
- Fuzzy clustering
Why?
- Are there subgroups of a disease? (e.g. multiple types of breast cancer, which require
different treatment)
Can we split the data meaningfully into different groups?
Do we find a pattern here?
Example data 2D, normally D is much larger (e.g. number of genes measured), but I can’t
draw that
There is no one correct answer. It’s not clear how many groups are here
Algorithm to put data into two groups
K-means clustering algorithm
Decide how many clusters you should have
This is not always straightforward
1. The algorithm starts by randomly assigning each of your observations to only one of
the k clusters.
