Machine Learning is the study of methods for programming computers to learn. Computers are applied to a wide range of tasks, and for most of these it is relatively easy for programmers to design and implement the necessary software. However, there are many tasks for which this is difficult or impossible. These can be divided into four general categories. First, there are problems for which there exist no human experts. For example, in modern automated manufacturing facilities, there is a need to predict machine failures before they occur by analyzing sensor readings. Because the machines are new, there are no human experts who can be interviewed by a programmer to provide the knowledge necessary to build a computer system. A machine learning system can study recorded data and subsequent machine failures and learn prediction rules. Second, there are problems where human experts exist, but where they are unable to explain their expertise. This is the case in many perceptual tasks, such as speech recognition, hand-writing recognition, and natural language understanding. Virtually all humans exhibit expert-level abilities on these tasks, but none of them can describe the detailed steps that they follow as they perform them. Fortunately, humans can provide machines with examples of the inputs and correct outputs for these tasks, so machine learning algorithms can learn to map the inputs to the outputs. Third, there are problems where phenomena are changing rapidly. In finance, for example, people would like to predict the future behavior of the stock market, of consumer purchases, or of exchange rates. These behaviors change frequently, so that even if a programmer could construct a good predictive computer program, it would need to be rewritten frequently. A learning program can relieve the programmer of this burden by constantly modifying and tuning a set of learned prediction rules. Fourth, there are applications that need to be customized for each computer user separately. Consider, for example, a program to filter unwanted electronic mail messages. Different users will need different filters. It is unreasonable to expect each user to program his or her own rules, and it is infeasible to provide every user with a software engineer to keep the rules up-to-date. A machine learning system can learn which mail messages the user rejects and maintain the filtering rules automatically. Machine learning addresses many of the same research questions as the fields of statistics, data mining, and psychology, but with differences of emphasis. Statistics focuses on understanding the phenomena that have generated the data, often with the goal of testing different hypotheses about those phenomena. Data mining seeks to find patterns in the data that are understandable by people. Psychological studies of human learning aspire to understand the mechanisms underlying the various learning behaviors exhibited by people (concept learning, skill acquisition, strategy change, etc.).

Machine learning

We will review some of the major results in random graphs and some of the more challenging open problems. We will cover algorithmic and structural questions. We will touch on newer models, including those related to the WWW.

Random graphs

Assessment of genomic DNA sequence variation and genotype calling in autotetraploids implies the ability to distinguish among five possible alternative allele copy number states. This study demonstrates the accuracy of genotyping-by-sequencing (GBS) of a large collection of autotetraploid potato cultivars using next-generation sequencing. It is still costly to reach sufficient read depths on a genome wide scale, across the cultivated gene pool. Therefore, we enriched cultivar-specific DNA sequencing libraries using an in-solution hybridisation method (SureSelect). This complexity reduction allowed to confine our study to 807 target genes distributed across the genomes of 83 tetraploid cultivars and one reference (DM 1–3 511). Indexed sequencing libraries were paired-end sequenced in 7 pools of 12 samples using Illumina HiSeq2000. After filtering and processing the raw sequence data, 12.4 Gigabases of high-quality sequence data was obtained, which mapped to 2.1 Mb of the potato reference genome, with a median average read depth of 63× per cultivar. We detected 129,156 sequence variants and genotyped the allele copy number of each variant for every cultivar. In this cultivar panel a variant density of 1 SNP/24 bp in exons and 1 SNP/15 bp in introns was obtained. The average minor allele frequency (MAF) of a variant was 0.14. Potato germplasm displayed a large number of relatively rare variants and/or haplotypes, with 61% of the variants having a MAF below 0.05. A very high average nucleotide diversity (p = 0.0107) was observed. Nucleotide diversity varied among potato chromosomes. Several genes under selection were identified. Genotyping-by-sequencing results, with allele copy number estimates, were validated with a KASP genotyping assay. This validation showed that read depths of ~60–80× can be used as a lower boundary for reliable assessment of allele copy number of sequence variants in autotetraploids. Genotypic data were associated with traits, and alleles strongly influencing maturity and flesh colour were identified.

/pdf/a-next-generation-sequencing-method-for-genotyping-by-1zm497w0y3.pdf

A Next-Generation Sequencing Method for Genotyping-by-Sequencing of Highly Heterozygous Autotetraploid Potato

Recent progress in mapping transcription factor (TF) binding regions can largely be credited to chromatin immunoprecipitation (ChIP) technologies. We compared strategies for mapping TF binding regions in mammalian cells using two different ChIP schemes: ChIP with DNA microarray analysis (ChIP-chip) and ChIP with DNA sequencing (ChIP-PET). We first investigated parameters central to obtaining robust ChIP-chip data sets by analyzing STAT1 targets in the ENCODE regions of the human genome, and then compared ChIP-chip to ChIP-PET. We devised methods for scoring and comparing results among various tiling arrays and examined parameters such as DNA microarray format, oligonucleotide length, hybridization conditions, and the use of competitor Cot-1 DNA. The best performance was achieved with high-density oligonucleotide arrays, oligonucleotides >/=50 bases (b), the presence of competitor Cot-1 DNA and hybridizations conducted in microfluidics stations. When target identification was evaluated as a function of array number, 80%-86% of targets were identified with three or more arrays. Comparison of ChIP-chip with ChIP-PET revealed strong agreement for the highest ranked targets with less overlap for the low ranked targets. With advantages and disadvantages unique to each approach, we found that ChIP-chip and ChIP-PET are frequently complementary in their relative abilities to detect STAT1 targets for the lower ranked targets; each method detected validated targets that were missed by the other method. The most comprehensive list of STAT1 binding regions is obtained by merging results from ChIP-chip and ChIP-sequencing. Overall, this study provides information for robust identification, scoring, and validation of TF targets using ChIP-based technologies.

/pdf/mapping-of-transcription-factor-binding-regions-in-mammalian-3bdy9emuo1.pdf

Mapping of transcription factor binding regions in mammalian cells by ChIP: Comparison of array- and sequencing-based technologies

An introduction to computer vision illumination and fixturing sensors image acquisition and representation fundamentals of digital image processing image analysis the segmentation problem 2-D shape description and recognition 3-D object representations robot programming and robot vision bin picking trends and aspirations.

Machine vision: automated visual inspection and robot vision

A recent development in microarray research entails the unbiased coverage, or tiling, of genomic DNA for the large-scale identification of transcribed sequences and regulatory elements. A central issue in designing tiling arrays is that of arriving at a single-copy tile path, as significant sequence cross-hybridization can result from the presence of non-unique probes on the array. Due to the fragmentation of genomic DNA caused by the widespread distribution of repetitive elements, the problem of obtaining adequate sequence coverage increases with the sizes of subsequence tiles that are to be included in the design. This becomes increasingly problematic when considering complex eukaryotic genomes that contain many thousands of interspersed repeats. The general problem of sequence tiling can be framed as finding an optimal partitioning of non-repetitive subsequences over a prescribed range of tile sizes, on a DNA sequence comprising repetitive and non-repetitive regions. Exact solutions to the tiling problem become computationally infeasible when applied to large genomes, but successive optimizations are developed that allow their practical implementation. These include an efficient method for determining the degree of similarity of many oligonucleotide sequences over large genomes, and two algorithms for finding an optimal tile path composed of longer sequence tiles. The first algorithm, a dynamic programming approach, finds an optimal tiling in linear time and space; the second applies a heuristic search to reduce the space complexity to a constant requirement. A Web resource has also been developed, accessible at http://tiling.gersteinlab.org, to generate optimal tile paths from user-provided DNA sequences.

/pdf/design-optimization-methods-for-genomic-dna-tiling-arrays-3mdr4lqsi1.pdf

Design optimization methods for genomic DNA tiling arrays

The recent explosion of genomic-scale protein interaction screens has made it possible to study protein interactions on a level of interactome and networks. In this chapter, we begin with an introduction of a novel approach that probabilistically combines multiple information sources to predict protein interactions in yeast. Specifically, Section 5.2 describes the sources of genomic features. Section 5.3 provides a basic tutorial on machine-learning approaches and

/pdf/protein-interaction-prediction-by-integrating-genomic-2fohpgxivt.pdf

Falk Schubert

Papers

Design optimization methods for genomic DNA tiling arrays

Protein Interaction Prediction by Integrating Genomic Features and Protein Interaction Network Analysis.