2008 Woodie Awards
Silenced gene research gains loud praise
By: Lisa Du
Issue date: 12/5/07 Section: News
Last update: 12/5/07 at 5:57 AM EST
Last update: 12/5/07 at 5:57 AM EST
- < prev Page 2 of 2
"Every imprinted gene that we have is going to have a characteristic DNA pattern for the female-marked ones and the male-marked imprinted genes," Jirtle said. "And if that is the case, we should be able to use computers to find those patterns and they should be uniquely present in genes that are imprinted."
He said two different machine-learning computer programs were used in identifying genes, and the overlap in the results was the 156 possible imprinted genes in the study.
"We used one existing technology and one we had developed in our own lab to do the predictions," said Hartemink, who wrote the algorithm for the program.
Jason Bosko, a senior in the Pratt School of Engineering who helped write the program Sparse Multinomial Logistic Regression, said finding the imprinted genes was similar to a classification problem.
The program had to compare existing information about 40 imprinted and 700 nonimprinted genes to locate unknown imprinted genes in the human genome, he added.
"You tell the computer some genes that are imprinted, and also some genes that are not imprinted, and you provide a set of descriptors, or features, for each of the genes," Hartemink said. "Then the computer considers features and combinations of features that are predictive of genes belonging to one class or the other. Eventually, you train the computer to distinguish between these classes by some complicated function of a subset of the features."
He said two different machine-learning computer programs were used in identifying genes, and the overlap in the results was the 156 possible imprinted genes in the study.
"We used one existing technology and one we had developed in our own lab to do the predictions," said Hartemink, who wrote the algorithm for the program.
Jason Bosko, a senior in the Pratt School of Engineering who helped write the program Sparse Multinomial Logistic Regression, said finding the imprinted genes was similar to a classification problem.
The program had to compare existing information about 40 imprinted and 700 nonimprinted genes to locate unknown imprinted genes in the human genome, he added.
"You tell the computer some genes that are imprinted, and also some genes that are not imprinted, and you provide a set of descriptors, or features, for each of the genes," Hartemink said. "Then the computer considers features and combinations of features that are predictive of genes belonging to one class or the other. Eventually, you train the computer to distinguish between these classes by some complicated function of a subset of the features."
Viewing Comments 1 - 1 of 1
April Eppler
posted 12/05/07 @ 3:36 PM EST
I have a six year old autistic son, so this is very interesting to me.
Post a Comment