Evaluating the accuracy and sensitivity of detecting minority HIV-1 populations by Illumina next-generation sequencing.

The accuracy and sensitivity of deep sequencing were assessed using viral standards (pNL4-3 and pLAI.2) of both DNA and RNA. The sequencing accuracy did not depend on the type of nucleic acid, but critically depended on the number of reads and threshold of sensitivity to minor viral populations. With coverage of more than 236 reads, the accuracy of viral RNA sequencing was equal to or exceeded 99.9%, with a sensitivity threshold to minor nucleotides of 20%. When the sensitivity threshold was below 1%, reduced accuracy dynamics were clearly visible even when the coverage was massive (more than 9.000 reads). It was found that the floating sensitivity threshold allowed the sequencing accuracy to be maintained at an acceptable level in cases of low coverage (less than 1.500-2.000) of reads. These results indicate the quality that can be expected with a specific number of reads and sensitivity threshold. Deep sequencing is a very powerful tool that can significantly improve the value of study results, but despite its superior performance, it should be used with caution regarding its sensitivity to minor populations below 1%.

[Machine learning study of DNA binding by transcription factors from the LacI family].

We studied 1372 LacI-family transcription factors and their 4484 DNA binding sites using machine learning algorithms and feature selection techniques. The Naive Bayes classifier and Logistic Regression were used to predict binding sites given transcription factor sequences and to classify factor-site pairs on binding and non-binding ones. Prediction accuracy was estimated using 10-fold cross-validation. Experiments showed that the best prediction of nucleotide densities at selected site positions is obtained using only a few key protein sequence positions. These positions are stably selected by the forward feature selection based on the mutual information of factor-site position pairs.