Research Project Proposal: Machine Learning algorithms ...
Transcript of Research Project Proposal: Machine Learning algorithms ...
![Page 1: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/1.jpg)
1
Research Project Proposal:
Machine Learning algorithms applied on RNA-seq data to classify subtypes of
Colorectal Cancer• Chiara Barbera
• Computer Science and Engineering
![Page 2: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/2.jpg)
2
Overview
✔ Research topic
✔ State of the art
- CRIS classification
✔ Our contribute
![Page 3: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/3.jpg)
3
Research TopicBackground information
![Page 4: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/4.jpg)
4
Key points
✔ Improve classification of subtypes of ColoRectal Cancer
✔ Use information on gene activity (gene expression)
✔ Cope with computational challenges
![Page 5: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/5.jpg)
5
What is Colorectal Cancer (CRC)
✔ Abnormal growth (polyps) in the colon rectum
✔ If cancerous, polyps can spread to lymph nodes and other organs1
✔ Heterogeneity of prognosis and therapy response
~90% 5-year survival rate
(localized)2
~15% 5-year survival rate
(distant)2
1) “What is colorectal cancer?.” https://www.cancer.org/cancer/colon-rectal-cancer/about/what-is-colorectal-cancer.html. Accessed: 2020-03-022.2) “Key statistics for colorectal cancer.” https://www.cancer.org/cancer/colon-rectal-cancer/about/key-statistics.html. Accessed: 2020-03-022.
Images from pngwave.com
![Page 6: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/6.jpg)
6
Genomics, epigenomics, genes
Images from pngwave.com; 1) “Talking glossary of genetic terms.” https://www.genome.gov/genetics-glossary . Accessed: 2020-03-027.2) “What is epigenetics?.” https://ghr.nlm.nih.gov/primer/howgeneswork/epigenome . Accessed: 2020-03-031.3) Number of genes from https://ghr.nlm.nih.gov/primer/basics/gene
✔ Gene1: portion of the DNA that encodes for a product (RNA or protein). Placed on
chromosomes, usually named with a short combination of letters (and possibly numbers)
✔ Gene expression1: measure of the activity of the genes
✔ Epigenome2: factors that modify how genes are expressed without modifying the DNA
20k – 25k Genes
in every person3
<1% Different sequences
of genes between individuals3
23 Human pairs of chromosomes
![Page 7: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/7.jpg)
7
From samples to subtypes
Images from pngwave.com
Samples
Prognosis
Therapy
Subtypes
Data extraction
MicroarrayNGS
Gene expression
CIMPCpG Island Methylator Phenotype
MSIMicroSatellite
Instability
Mutations
Classification
NTPkTSPSVMNeural Network...
![Page 8: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/8.jpg)
8
Great, but...
Images from pngwave.com
Unbalanced dataset Variety of platforms
0
500
1000
1500
2000
2500
3000
3500
Few data, too many features Curse of dimensionalityVariety of normalizations
FPKMRPKM TPM
Ambiguity of data
![Page 9: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/9.jpg)
9
Platforms
Images from pngwave.com
MICROARRAY1
✔ Older✔ Uses fluorochrome to mark
binded sequences on a chip✔ Can use only a limited set of
genes on the chip
NGS2
✔ More recent✔ Parallel sequencing✔ Can sequence whole genome✔ RNA-seq, DNA-seq and other
1) “Talking glossary of genetic terms.” https://www.genome.gov/genetics-glossary . Accessed: 2020-03-027.2) “Definition of next-generation sequencing.” https://www.cancer.gov/publications/dictionaries/genetics-dictionary/def/next-generation-sequencing . Accessed: 2020-03-024.
![Page 10: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/10.jpg)
10
State of the artHow did researchers face these issues?
![Page 11: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/11.jpg)
11
Consensus Molecular Subtype Classification
SUBTYPE SOME CHARACTERISTICS DISTRIBUTION OF CMS
CMS1 MSI 14 %
CMS2 Chromosome Instability 37 %
CMS3 Methabolic disregulation 13 %
CMS4 Stromal influence 23 %
87% of samples classified
✔ 4 subtypes✔ Aggregated 6 different classifications and 18 datasets (Network approach + Markov Clustering Algorithm) and Random Forest Classifier
✗ 22 % = non-consensus samples, classified through a probability threshold
Ambiguity of data
J. Guinney, R. Dienstmann, X. Wang, A. Reyniès, A. Schlicker, C. Soneson, L. Marisa, P. Roepman, G. Nyamundanda, P. Angelino, B. Bot, J. Morris, I. Simon, S. Gerster, E. Fessler, F. De Sousa E Melo, E. Missiaglia, H. Ramay, D. Barras, and S. Tejpar, “The consensus molecular subtypes of colorectal cancer,” Nature Medicine, vol. 21, pp. 1350–1356, 2015.
![Page 12: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/12.jpg)
12
Study for chemotherapy response✔ Differences in prognosis and chemotherapy response according to subtype
✗ Only 71% of classified samples
✗ Few samples respecting therapy requirements for the analysis
Ambiguity of dataMSI STATUS CIMPBRAF
MUTATIONKRAS
MUTATION
SUBTYPE 1 INSTABLE + + -
SUBTYPE 2 STABLE + + -
SUBTYPE 3 STABLE - - +
SUBTYPE 4 STABLE - - -
SUBTYPE 5 INSTABLE - - -
0
500
1000
1500
2000
2500
3000
3500
Few samples
O. Murcia, M. Juárez, M. Rodríguez-Soler, E. Hernández-Illán, M. Giner-Calabuig, M. Alustiza, C. Egoavil, A. Castillejo, C. Alenda, V. Barberá, C. Mangas-Sanjuan, A. Yuste, L. Bujanda, J. Clofent, M. Andreu, A. Castells, X. Llor, P. Zapater, and R. Jover, “Colorectal cancer molecular classification using braf, kras, microsatellite instability and cimp status: Prognostic implications and response to chemotherapy,” Plos One, vol. 13, p. e0203051, 2018.
![Page 13: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/13.jpg)
13
Feature Specific Quantile Normalization (FSQN)
✔ Data on Breast and Colorectal cancer (Consensus Molecular Subtype)
✔ Train on microarray, validation on RNA-seq data
✔ Several normalization procedure. The FSQN showed the highest performances
Variety of platformsVariety of normalizations
FPKMRPKM TPM
J. Franks, G. Cai, and M. Whitfield, “Feature specific quantile normalization enables cross-platform classification of molecular subtypes using gene expression data,” Bioinformatics (Oxford, England), vol. 34, p. 1868–1874, 2018.
![Page 14: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/14.jpg)
14
Binary and Pan classifier
✔ First Classifier: Normal VS cancerous cells
✔ Second Classifier: for 21 types of tumors
✔ Selected different set of genes and applied different algorithms✔ Neural Network✔ Linear Support Vector Machine✔ Radial Basis Function Support Vector Machine✔ K-Nearest Neighbours✔ Random Forest
✗ Some samples were misclassified by pan classifier (near regions)Ambiguity of data
0
500
1000
1500
2000
2500
3000
3500
Few samples, too many features
B.-H. Kim, K. Yu, and P. Lee, “Cancer classification of single cell gene expression data by neural network,”Bioinformatics (Oxford, England), vol. 36, p. 1–7, 2019.
![Page 15: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/15.jpg)
15
Our objective
Single-sample classifier for CRIS1 subtypes of Colorectal Cancer (CRC)
Why ?
✔ Improve the current approaches and results
✔ Only RNA-seq data (NGS)
✔ Necessary step towards a clinical application
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 16: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/16.jpg)
16
Our starting pointThe CRIS Classification
![Page 17: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/17.jpg)
17
CRIS Classification
SUBTYPE SOME CHARACTERISTICS
CRIS-A Enriched for MSI or KRAS gene mutation
CRIS-B Poor prognosis
CRIS-C Sensitivity to EFGR inibitors (responsive to cetuximab)
CRIS-D IGF2 gene overexpression
CRIS-E TP53 gene mutation
✔ 5 CRIS subtypes (ColoRectal Intrinsic Subtypes)✔ Training on microarray, testing on both microarray and RNA-seq
Variety of platforms
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 18: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/18.jpg)
18
Machine learning analysis
1) Non-negative Matrix Factorization (NMF) on PDX (Patient Derived Xenografts) data to identify 5 clusters (subtypes)
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 19: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/19.jpg)
19
Machine learning analysis
1) Non-negative Matrix Factorization (NMF) on PDX (Patient Derived Xenografts) data to identify 5 clusters (subtypes)
2) Definition of templates of each subtype, based on NMF
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 20: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/20.jpg)
20
Machine learning analysis
1) Non-negative Matrix Factorization (NMF) on PDX (Patient Derived Xenografts) data to identify 5 clusters (subtypes)
2) Definition of templates of each subtype, based on NMF
3) NTP (Nearest Template Prediction) classification
✔ On dataset (microarray + RNA-seq data)
✔ Good accuracy
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 21: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/21.jpg)
21
Machine learning analysis
1) Non-negative Matrix Factorization (NMF) on PDX (Patient Derived Xenografts) data to identify 5 clusters (subtypes)
2) Definition of templates of each subtype, based on NMF
3) NTP (Nearest Template Prediction) classification
✔ On dataset (microarray + RNA-seq data)
✔ Good accuracy
4) Attempt of single-sample classifier through kTSP (k Top Scoring Pairs)
✗ Suboptimal performances
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 22: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/22.jpg)
22
Machine learning analysis
1) Non-negative Matrix Factorization (NMF) on PDX (Patient Derived Xenografts) data to identify 5 clusters (subtypes)
2) Definition of templates of each subtype, based on NMF
3) NTP classification
✔ On dataset (microarray + RNA-seq data)
✔ Good accuracy
4) Attempt of single-sample classifier through kTSP (k Top Scoring Pairs)
✗ Suboptimal performances
0
500
1000
1500
2000
2500
3000
3500
Few samples, too many features
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 23: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/23.jpg)
23
Results of NTP analysis
16 Unbalanced dataset
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 24: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/24.jpg)
24
Pros and Cons of CRIS
✔ Removed stromal influence
✔ Good performances for the dataset classifier
✔ Cross – platform (Microarray and RNA-seq data)
✗ Single–sample classifier is suboptimal
✗ Microarray do not provide as many features as RNA-seq data
✗ Instability of classification for some samples due to dataset composition dependency (z-score)
✗ Some genes may still be stromal
Ambiguity of data
1) C. Isella, F. Brundu, S. E. Bellomo, F. Galimi, E. Zanella, R. Porporato, C. Petti, A. Fiori, F. Orzan, R. Senetta, C. Boccaccio, E. Ficarra, L. Marchionni, L. Trusolino, E. Medico, and A. Bertotti, “Selective analysis of cancer-cell intrinsic transcriptional traits defines novel clinically relevant subtypes of colorectal cancer,” Nature Communications, vol. 8, p. 15107, 2017.
![Page 25: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/25.jpg)
25
Our contributionWhat we want to do
![Page 26: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/26.jpg)
26
What do we want to investigate?
✔ Is CRIS classification stable on RNA-seq only?
![Page 27: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/27.jpg)
27
What do we want to investigate?
✔ Is CRIS classification stable on RNA-seq only?
✔ Are there any new features (more meaningful) on RNA-seq data?
![Page 28: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/28.jpg)
28
What do we want to investigate?
✔ Is CRIS classification stable on RNA-seq only?
✔ Are there any new features (more meaningful) on RNA-seq data?
✔ Can we obtain a single-sample classifier with high accuracy?
![Page 29: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/29.jpg)
29
What do we want to investigate?
✔ Is CRIS classification stable on RNA-seq only?
✔ Are there any new features (more meaningful) on RNA-seq data?
✔ Can we obtain a single-sample classifier with high accuracy?
✔ If new features are selected, what is their biological meaning?
![Page 30: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/30.jpg)
30
What do we want to investigate?
✔ Is CRIS classification stable on RNA-seq only?
✔ Are there any new features (more meaningful) on RNA-seq data?
✔ Can we obtain a single-sample classifier with high accuracy?
✔ If new features are selected, what is their biological meaning?
✔ Are there other subtypes for CRIS? If so, does the accuracy improve?
![Page 31: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/31.jpg)
31
How will we face the issues?
✔ Curse of dimensionality and small dataset?
Apply feature selection to select only the important features
✔ Heterogeneity of platform and data? Only RNA-seq data and uniform type of normalization
✔ Unbalanced dataset? Sample selection
![Page 32: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/32.jpg)
32
What is new in the research?
✔ Application of CRIS on RNA-seq data only because they have more features
✔ No single-sample classifier is currently available
✔ Single-sample classifier development is necessary for clinical application
✔ Possibly develop a new algorithm for feature selection
✔ Possibly hybrid approaches (starting with unsupervised) for classification
![Page 33: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/33.jpg)
33
Data and tools
RNA-seq data from
✔ More features
✔ More precise
✔ Current trend technology
TCGA (Public)
PDX (proprietary, Candiolo)
![Page 34: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/34.jpg)
34
Steps of our work1) Collection of data (TCGA RNA-Seq data + PDX RNA-seq data)
![Page 35: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/35.jpg)
35
Steps of our work1) Collection of data (TCGA RNA-Seq data + PDX RNA-seq data)
2) Replication of CRIS NTP and kTSP classifier on these data
![Page 36: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/36.jpg)
36
Steps of our work1) Collection of data (TCGA RNA-Seq data + PDX RNA-seq data)
2) Replication of CRIS NTP and kTSP classifiers on these data
3) Study of possible alternative classifiers
1) Feature selection and sample selection
2) Execution of classifiers
![Page 37: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/37.jpg)
37
Steps of our work1) Collection of data (TCGA RNA-Seq data + PDX RNA-seq data)
2) Replication of CRIS NTP and kTSP classifiers on these data
3) Study of possible alternative classifiers
1) Feature selection and sample selection
2) Execution of classifiers
4) Performance comparison
1) CRIS classification of Isella et al.
2) CRIS classification on RNA-seq data only
3) Alternative classifiers
![Page 38: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/38.jpg)
38
Steps of our work1) Collection of data (TCGA RNA-Seq data + PDX RNA-seq data)
2) Replication of CRIS NTP and kTSP classifiers on these data
3) Study of possible alternative classifiers
1) Feature selection and sample selection
2) Execution of classifiers
4) Performance comparison
1) CRIS classification of Isella et al.
2) CRIS classification on RNA-seq data only
3) Alternative classifiers
5) Clinical and biological validation
![Page 39: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/39.jpg)
39
Conclusions
1) Several issues to take into account
![Page 40: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/40.jpg)
40
Conclusions
1) Several issues to take into account
2) Other studies coped with these issues and reached relevant results on classification
and response to therapy
![Page 41: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/41.jpg)
41
Conclusions
1) Several issues to take into account
2) Other studies coped with these issues and reached relevant results on classification and
response to therapy
3) We want to improve the results of the CRIS classification by applying alternative
classifiers on RNA-seq data only
![Page 42: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/42.jpg)
42
Conclusions
1) Several issues to take into account
2) Other studies coped with these issues and reached relevant results on classification and
response to therapy
3) We want to improve the results of the CRIS classification by applying alternative classifiers
on RNA-seq data only (new feature selection and/or classification algorithm)
4) We would like to have a single-sample classifier with acceptable accuracy
![Page 43: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/43.jpg)
43
Questions?
![Page 44: Research Project Proposal: Machine Learning algorithms ...](https://reader030.fdocuments.us/reader030/viewer/2022032610/6239bee773a5692fb43ef7ae/html5/thumbnails/44.jpg)
44
Thank you for your attention!