Limits...
DISIS: prediction of drug response through an iterative sure independence screening.

Fang Y, Qin Y, Zhang N, Wang J, Wang H, Zheng X - PLoS ONE (2015)

Bottom Line: Prediction of drug response based on genomic alterations is an important task in the research of personalized medicine.Current elastic net model utilized a sure independence screening to select relevant genomic features with drug response, but it may neglect the combination effect of some marginally weak features.For each drug in CCLE, we selected up to 40 features including gene expressions, mutation and copy number alterations of cancer-related genes, and some of them are significantly strong features but showing weak marginal correlation with drug response vector.

View Article: PubMed Central - PubMed

Affiliation: Department of Mathematics, Shanghai Normal University, Shanghai, China.

ABSTRACT
Prediction of drug response based on genomic alterations is an important task in the research of personalized medicine. Current elastic net model utilized a sure independence screening to select relevant genomic features with drug response, but it may neglect the combination effect of some marginally weak features. In this work, we applied an iterative sure independence screening scheme to select drug response relevant features from the Cancer Cell Line Encyclopedia (CCLE) dataset. For each drug in CCLE, we selected up to 40 features including gene expressions, mutation and copy number alterations of cancer-related genes, and some of them are significantly strong features but showing weak marginal correlation with drug response vector. Lasso regression based on the selected features showed that our prediction accuracies are higher than those by elastic net regression for most drugs.

No MeSH data available.


Related in: MedlinePlus

Scatter plots of the true and predicted sensitivities for some drugs.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4368776&req=5

pone.0120408.g004: Scatter plots of the true and predicted sensitivities for some drugs.

Mentions: Scatter plots of the observed and predicted responses for some typical drugs are shown as Fig. 4. We could conclude from these examples that the resulting correlation was fairly reasonable and not overestimated by a few outliers. Moreover, our predictions were in great consistence with those by ENR model, given the overall correlation of 0.81 (Fig. 5). In particular, if we discarded the only one outlier, Nilotinib, the overall Pearson correlation increased from 0.81 to 0.94. As is known, Nilotinib is a special compound for treating chronic myelogenous leukemia (CML) [26], which was successfully selected as the strongest feature for sensitivity of Nilotinib according to the CCLE paper [9]. So it is reasonable that this top feature dominated the model building and prediction, and brought a high prediction correlation by ENR. While we did not take this lineage information into consideration since we focused only on mining genomic information to explain the drug sensitivity. Except this only outlier, ISIS brought much higher predicted correlations using fewer features than ENR for most drugs, which confirmed the efficiency of ISIS in feature selection.


DISIS: prediction of drug response through an iterative sure independence screening.

Fang Y, Qin Y, Zhang N, Wang J, Wang H, Zheng X - PLoS ONE (2015)

Scatter plots of the true and predicted sensitivities for some drugs.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4368776&req=5

pone.0120408.g004: Scatter plots of the true and predicted sensitivities for some drugs.
Mentions: Scatter plots of the observed and predicted responses for some typical drugs are shown as Fig. 4. We could conclude from these examples that the resulting correlation was fairly reasonable and not overestimated by a few outliers. Moreover, our predictions were in great consistence with those by ENR model, given the overall correlation of 0.81 (Fig. 5). In particular, if we discarded the only one outlier, Nilotinib, the overall Pearson correlation increased from 0.81 to 0.94. As is known, Nilotinib is a special compound for treating chronic myelogenous leukemia (CML) [26], which was successfully selected as the strongest feature for sensitivity of Nilotinib according to the CCLE paper [9]. So it is reasonable that this top feature dominated the model building and prediction, and brought a high prediction correlation by ENR. While we did not take this lineage information into consideration since we focused only on mining genomic information to explain the drug sensitivity. Except this only outlier, ISIS brought much higher predicted correlations using fewer features than ENR for most drugs, which confirmed the efficiency of ISIS in feature selection.

Bottom Line: Prediction of drug response based on genomic alterations is an important task in the research of personalized medicine.Current elastic net model utilized a sure independence screening to select relevant genomic features with drug response, but it may neglect the combination effect of some marginally weak features.For each drug in CCLE, we selected up to 40 features including gene expressions, mutation and copy number alterations of cancer-related genes, and some of them are significantly strong features but showing weak marginal correlation with drug response vector.

View Article: PubMed Central - PubMed

Affiliation: Department of Mathematics, Shanghai Normal University, Shanghai, China.

ABSTRACT
Prediction of drug response based on genomic alterations is an important task in the research of personalized medicine. Current elastic net model utilized a sure independence screening to select relevant genomic features with drug response, but it may neglect the combination effect of some marginally weak features. In this work, we applied an iterative sure independence screening scheme to select drug response relevant features from the Cancer Cell Line Encyclopedia (CCLE) dataset. For each drug in CCLE, we selected up to 40 features including gene expressions, mutation and copy number alterations of cancer-related genes, and some of them are significantly strong features but showing weak marginal correlation with drug response vector. Lasso regression based on the selected features showed that our prediction accuracies are higher than those by elastic net regression for most drugs.

No MeSH data available.


Related in: MedlinePlus