Optimization of signal-to-noise ratio for efficient microarray probe design

NNDB: An Expanded Database of Nearest Neighbor Parameters for Predicting Stability of Nucleic Acid Secondary Structures

Nearest neighbor thermodynamic parameters are widely used for RNA and DNA secondary structure prediction and to model thermodynamic ensembles of secondary structures. The Nearest Neighbor Database (NNDB) is a freely available web resource (https://rna.urmc.rochester.edu/NNDB) that provides the functional forms, parameter values, and example calculations. The NNDB provides the 1999 and 2004 set of RNA folding nearest neighbor parameters. We expanded the database to include a set of DNA parameters and a set of RNA parameters that includes m6A in addition to the canonical RNA nucleobases. The site was redesigned using the Quarto open-source publishing system. A downloadable PDF version of the complete resource and downloadable sets of nearest neighbor parameters are available.

FREQ-Seq2: a method for precise high-throughput combinatorial quantification of allele frequencies

The accurate determination of allele frequencies is crucially important across a wide range of problems in genetics, such as developing population genetic models, making inferences from genome-wide association studies, determining genetic risk for diseases, as well as other scientific and medical applications. Furthermore, understanding how allele frequencies change over time in populations is central to ascertaining their evolutionary dynamics. We present a precise, efficient, and economical method (FREQ-Seq2) for quantifying the relative frequencies of different alleles at loci of interest in mixed population samples. Through the creative use of paired barcode sequences, we exponentially increased the throughput of the original FREQ-Seq method from 48 to 2,304 samples. FREQ-Seq2 can be targeted to specific genomic regions of interest, which are amplified using universal barcoded adapters to generate Illumina sequencing libraries. Our enhanced method, available as a kit along with open-source software for analyzing sequenced libraries, enables the detection and removal of errors that are undetectable in the original FREQ-Seq method as well as other conventional methods for allele frequency quantification. Finally, we validated the performance of our sequencing-based approach with a highly multiplexed set of control samples as well as a competitive evolution experiment in Escherichia coli and compare the latter to estimates derived from manual colony counting. Our analyses demonstrate that FREQ-Seq2 is flexible, inexpensive, and produces large amounts of data with low error, low noise, and desirable statistical properties. In summary, FREQ-Seq2 is a powerful method for quantifying allele frequency that provides a versatile approach for profiling mixed populations.

A decision-theoretic approach to the evaluation of machine learning algorithms in computational drug discovery

Motivation

Artificial intelligence, trained via machine learning (e.g. neural nets, random forests) or computational statistical algorithms (e.g. support vector machines, ridge regression), holds much promise for the improvement of small-molecule drug discovery. However, small-molecule structure-activity data are high dimensional with low signal-to-noise ratios and proper validation of predictive methods is difficult. It is poorly understood which, if any, of the currently available machine learning algorithms will best predict new candidate drugs.

Results

The quantile-activity bootstrap is proposed as a new model validation framework using quantile splits on the activity distribution function to construct training and testing sets. In addition, we propose two novel rank-based loss functions which penalize only the out-of-sample predicted ranks of high-activity molecules. The combination of these methods was used to assess the performance of neural nets, random forests, support vector machines (regression) and ridge regression applied to 25 diverse high-quality structure-activity datasets publicly available on ChEMBL. Model validation based on random partitioning of available data favours models that overfit and ‘memorize’ the training set, namely random forests and deep neural nets. Partitioning based on quantiles of the activity distribution correctly penalizes extrapolation of models onto structurally different molecules outside of the training data. Simpler, traditional statistical methods such as ridge regression can outperform state-of-the-art machine learning methods in this setting. In addition, our new rank-based loss functions give considerably different results from mean squared error highlighting the necessity to define model optimality with respect to the decision task at hand.

Availability and implementation

All software and data are available as Jupyter notebooks found at https://github.com/owatson/QuantileBootstrap.

Supplementary information

Supplementary data are available at Bioinformatics online.

Ultraconserved element (UCE) probe set design: Base genome and initial design parameters critical for optimization

Targeted capture and enrichment approaches have proven effective for phylogenetic study. Ultraconserved elements (UCEs) in particular have exhibited great utility for phylogenomic analyses, with the software package phyluce being among the most utilized pipelines for UCE phylogenomics, including probe design. Despite the success of UCEs, it is becoming increasing apparent that diverse lineages require probe sets tailored to focal taxa in order to improve locus recovery. However, factors affecting probe design and methods for optimizing probe sets to focal taxa remain underexplored. Here, we use newly available beetle (Coleoptera) genomic resources to investigate factors affecting UCE probe set design using phyluce. In particular, we explore the effects of stringency during initial design steps, as well as base genome choice on resulting probe sets and locus recovery. We found that both base genome choice and initial bait design stringency parameters greatly alter the number of resultant probes included in final probe sets and strongly affect the number of loci detected and recovered during in silico testing of these probe sets. In addition, we identify attributes of base genomes that correlated with high performance in probe design. Ultimately, we provide a recommended workflow for using phyluce to design an optimized UCE probe set that will work across a targeted lineage, and use our findings to develop a new, open-source UCE probe set for beetles of the suborder Adephaga.

Sequence characteristics define trade-offs between on-target and genome-wide off-target hybridization of oligoprobes

Off-target oligoprobe's interaction with partially complementary nucleotide sequences represents a problem for many bio-techniques. The goal of the study was to identify oligoprobe sequence characteristics that control the ratio between on-target and off-target hybridization. To understand the complex interplay between specific and genome-wide off-target (cross-hybridization) signals, we analyzed a database derived from genomic comparison hybridization experiments performed with an Affymetrix tiling array. The database included two types of probes with signals derived from (i) a combination of specific signal and cross-hybridization and (ii) genomic cross-hybridization only. All probes from the database were grouped into bins according to their sequence characteristics, where both hybridization signals were averaged separately. For selection of specific probes, we analyzed the following sequence characteristics: vulnerability to self-folding, nucleotide composition bias, numbers of G nucleotides and GGG-blocks, and occurrence of probe's k-mers in the human genome. Increases in bin ranges for these characteristics are simultaneously accompanied by a decrease in hybridization specificity-the ratio between specific and cross-hybridization signals. However, both averaged hybridization signals exhibit growing trends along with an increase of probes' binding energy, where the hybridization specific signal increases significantly faster in comparison to the cross-hybridization. The same trend is evident for the S function, which serves as a combined evaluation of probe binding energy and occurrence of probe's k-mers in the genome. Application of S allows extracting a larger number of specific probes, as compared to using only binding energy. Thus, we showed that high values of specific and cross-hybridization signals are not mutually exclusive for probes with high values of binding energy and S. In this study, the application of a new set of sequence characteristics allows detection of probes that are highly specific to their targets for array design and other bio-techniques that require selection of specific probes.