Eugene Davydov

Computational efforts to identify functional elements within genomes leverage comparative sequence information by looking for regions that exhibit evidence of selective constraint. One way of detecting constrained elements is to follow a bottom-up approach by computing constraint scores for individual positions of a multiple alignment and then defining constrained elements as segments of contiguous, highly scoring nucleotide positions. Here we present GERP++, a new tool that uses maximum likelihood evolutionary rate estimation for position-specific scoring and, in contrast to previous bottom-up methods, a novel dynamic programming approach to subsequently define constrained elements. GERP++ evaluates a richer set of candidate element breakpoints and ranks them based on statistical significance, eliminating the need for biased heuristic extension techniques. Using GERP++ we identify over 1.3 million constrained elements spanning over 7% of the human genome. We predict a higher fraction than earlier estimates largely due to the annotation of longer constrained elements, which improves one to one correspondence between predicted elements with known functional sequences. GERP++ is an efficient and effective tool to provide both nucleotide- and element-level constraint scores within deep multiple sequence alignments.

To compare entire genomes from different species, biologists increasingly need alignment methods that are efficient enough to handle long sequences, and accurate enough to correctly align the conserved biological features between distant species. We present LAGAN, a system for rapid global alignment of two homologous genomic sequences, and Multi-LAGAN, a system for multiple global alignment of genomic sequences. We tested our systems on a data set consisting of greater than 12 Mb of high-quality sequence from 12 vertebrate species. All the sequence was derived from the genomic region orthologous to an approximately 1.5-Mb region on human chromosome 7q31.3. We found that both LAGAN and Multi-LAGAN compare favorably with other leading alignment methods in correctly aligning protein-coding exons, especially between distant homologs such as human and chicken, or human and fugu. Multi-LAGAN produced the most accurate alignments, while requiring just 75 minutes on a personal computer to obtain the multiple alignment of all 12 sequences. Multi-LAGAN is a practical method for generating multiple alignments of long genomic sequences at any evolutionary distance. Our systems are publicly available at http://lagan.stanford.edu.

Predicting ad click-through rates (CTR) is a massive-scale learning problem that is central to the multi-billion dollar online advertising industry. We present a selection of case studies and topics drawn from recent experiments in the setting of a deployed CTR prediction system. These include improvements in the context of traditional supervised learning based on an FTRL-Proximal online learning algorithm (which has excellent sparsity and convergence properties) and the use of per-coordinate learning rates.

Machine learning offers a fantastically powerful toolkit for building useful complex prediction systems quickly. This paper argues it is dangerous to think of these quick wins as coming for free. Using the software engineering framework of technical debt, we find it is common to incur massive ongoing maintenance costs in real-world ML systems. We explore several ML-specific risk factors to account for in system design. These include boundary erosion, entanglement, hidden feedback loops, undeclared consumers, data dependencies, configuration issues, changes in the external world, and a variety of system-level anti-patterns.

Machine learning offers a fantastically powerful toolkit for building complex sys-tems quickly. This paper argues that it is dangerous to think of these quick wins as coming for free. Using the framework of technical debt, we note that it is re-markably easy to incur massive ongoing maintenance costs at the system level when applying machine learning. The goal of this paper is highlight several ma-chine learning specific risk factors and design patterns to be avoided or refactored where possible. These include boundary erosion, entanglement, hidden feedback loops, undeclared consumers, data dependencies, changes in the external world, and a variety of system-level anti-patterns. 1 Machine Learning and Complex Systems Real world software engineers are often faced with the challenge of moving quickly to ship new products or services, which can lead to a dilemma between speed of execution and quality of en-gineering. The concept of technical debt was first introduced by Ward Cunningham in 1992 as a way to help quantify the cost of such decisions. Like incurring fiscal debt, there are often sound