How machine learning helps cancer research
Evelina Gabasova
@evelgab
MRC Cancer Unit, University of Cambridge
DNA
|
|
Cost of whole genome sequencing
Sequencing data
DNA and genes
Cancer
BRCA1 and BRCA2 are chromosome guardians
|
|
Clustering |
|
ExampleClustering wholesale customers440 wholesale customers
Annual spending on
|
|
Methods for clustering
- k-means clustering
- hierarchical clustering
- spectral clustering
- Gaussian mixture model and other probabilistic methods
- ...
Visualisation of high-dimensional data
Principal component analysis
Clustering cancer data
Genes instead of customers
Gene expression instead of spending on products
Conventional medicine
Precision medicine
Clustering in cancer research
TCGA breast cancer
Clustering 368 tumour samples based on expression of 648 genes.
Integrative clustering
Integrative clustering
Collaborative filtering |
|
Example
The Netflix prize
User |
Film 1 |
Film 2 |
Film 3 |
Film 4 |
... |
Film 1000 |
|---|---|---|---|---|---|---|
Alice |
5 |
2 |
x |
x |
... |
4 |
Bob |
x |
1 |
2 |
x |
... |
2 |
Carol |
2 |
x |
3 |
? |
... |
3 |
... |
... |
... |
... |
... |
... |
... |
Zoe |
x |
5 |
4 |
5 |
... |
x |
Matrix factorization
Collaborative filtering of cancer data
Patients instead of users
DNA mutations instead of film ratings
Mutational signatures
Patient |
C/A |
C/G |
C/T |
T/A |
T/C |
T/G |
|---|---|---|---|---|---|---|
Alice |
5 |
2 |
0 |
0 |
3 |
4 |
Bob |
0 |
1 |
2 |
0 |
0 |
2 |
Carol |
2 |
0 |
3 |
0 |
1 |
3 |
... |
... |
... |
... |
... |
... |
... |
Zoe |
0 |
5 |
4 |
5 |
2 |
0 |
Matrix factorization to identify features
 |
|
Matrix factorization to identify features
Matrix factorization to identify features
Proving system stability |
|
Theorem proving
SAT
(A ∨ ¬B ) ∧ (¬ A ∨ B)
A = true
B = true
Theorem proving
Satisfiability Modulo Theories (SMT)
(A ∨ ¬B ) ∧ (¬ A ∨ B)
((a > 3) ∨ (b < 1)) ∧ ((a < 5) ∨ (b = 0))
a = 4
b = 0
Software verification
Z3 theorem prover
| Preconditions | ||
| Postconditions | ||
| Loop conditions | ➜ | SMT formulas |
| Assertions | ||
| ... |
Software verification
Spec#
Software verification
Spec#
Proving stability of biological processes
|
Proteins | Genes | ➜ Variables | Receptors | … |
| ➜ | v + 1 | if v < T(v) | |
| v | ➜ | v | if v = T(v) |
| ➜ | v - 1 | if v > T(v) |
Chronic myeloid leukemia
Proving stability of biological systems
Chronic myeloid leukemia
Machine learning is not just
for targeted advertising
or algorithmic trading
 |
@evelgab |
 |
evelina@evelinag.com |
 |
github.com/evelinag |
| evelinag.com |
Links
- MRC Cancer Unit, University of Cambridge
- Big Data: Astronomical or Genomical?
- Signatures of mutational processes in cancer
- Z3 prover
- Proving stabilization of biological systems
- Bio Model Analyser
- Drug target optimization in Chronic Myeloid Leukemia using innovative computational platform
 |
 |