Hit Classification Research Articles

SCORCH: Improving structure-based virtual screening with machine learning classifiers, data augmentation, and uncertainty estimation

IntroductionThe discovery of a new drug is a costly and lengthy endeavour. The computational prediction of which small molecules can bind to a protein target can accelerate this process if the predictions are fast and accurate enough. Recent machine-learning scoring functions re-evaluate the output of molecular docking to achieve more accurate predictions. However, previous scoring functions were trained on crystalised protein-ligand complexes and datasets of decoys. The limited availability of crystal structures and biases in the decoy datasets can lower the performance of scoring functions. ObjectivesTo address key limitations of previous scoring functions and thus improve the predictive performance of structure-based virtual screening. MethodsA novel machine-learning scoring function was created, named SCORCH (Scoring COnsensus for RMSD-based Classification of Hits). To develop SCORCH, training data is augmented by considering multiple ligand poses and labelling poses based on their RMSD from the native pose. Decoy bias is addressed by generating property-matched decoys for each ligand and using the same methodology for preparing and docking decoys and ligands. A consensus of 3 different machine learning approaches is also used to improve performance. ResultsWe find that multi-pose augmentation in SCORCH improves its docking power and screening power on independent benchmark datasets. SCORCH outperforms an equivalent scoring function trained on single poses, with a 1 % enrichment factor (EF) of 13.78 vs. 10.86 on 18 DEKOIS 2.0 targets and a mean native pose rank of 5.9 vs 30.4 on CSAR 2014. Additionally, SCORCH outperforms widely used scoring functions in virtual screening and pose prediction on independent benchmark datasets. ConclusionBy rationally addressing key limitations of previous scoring functions, SCORCH improves the performance of virtual screening. SCORCH also provides an estimate of its uncertainty, which can help reduce the cost and time required for drug discovery.

Deep Learning applied to hit classification for the BESIII drift chamber

Drift chamber is the main tracking detector for high energy physics experiment BESIII. Due to the high luminosity and high beam intensity, the BESIII drift chamber is suffered from the beam background and electronics noise which represent a computing challenge to the reconstruction software. Deep learning developments in the last few years have shown tremendous improvements in the analysis of data especially for object classification. Here we present a first study of deep learning architectures applied to the real data of BESIII drift chamber to accomplish the hit classification of the background and signal.

Structural shift of gut microbiota during chemo-preventive effects of epigallocatechin gallate on colorectal carcinogenesis in mice.

AIMTo investigate the effect of epigallocatechin gallate (EGCG) on structural changes of gut microbiota in colorectal carcinogenesis.METHODSAn azoxymethane (AOM)/dextran sodium sulfate (DSS)-induced colitis mouse model was established. Forty-two female FVB/N mice were randomly divided into the following three groups: group 1 (10 mice, negative control) was treated with vehicle, group 2 (16 mice, positive control) was treated with AOM plus vehicle, and group 3 (16 mice, EG) was treated with AOM plus EGCG. For aberrant crypt foci (ACF) evaluation, the colons were rapidly took out after sacrifice, rinsed with saline, opened longitudinally, laid flat on a polystyrene board, and fixed with 10% buffered formaldehyde solution before being stained with 0.2% methylene blue in saline. For tumor evaluation, the colon was macroscopically inspected and photographed, then the total number of tumors was enumerated and tumor size measured. For histological examination, the fixed tissues were paraffin-embedded and sectioned at 5 mm thickness. Microbial genomic DNA was extracted from fecal and intestinal content samples using a commercial kit. The V4 hypervariable regions of 16S rRNA were PCR-amplified with the barcoded fusion primers. Using the best hit classification option, the sequences from each sample were aligned to the RDP 16S rRNA training set to classify the taxonomic abundance in QIIME. Statistical analyses were then performed.RESULTSTreatment of mice with 1% EGCG caused a significant decrease in the mean number of ACF per mouse, when compared with the model mice treated with AOM/DSS (5.38 ± 4.24 vs 13.13 ± 3.02, P < 0.01). Compared with the positive control group, 1% EGCG treatment dependently decreased tumor load per mouse by 85% (33.96 ± 6.10 vs 2.96 ± 2.86, respectively, P < 0.01). All revealed that EGCG could inhibit colon carcinogenesis by decreasing the number of precancerous lesions as well as solid tumors, with reduced tumor load and delayed histological progression of CRC. During the cancerization, the diversity of gut microbiota increased, potential carcinogenic bacteria such as Bacteroides were enriched, and the abundance of butyrate-producing bacteria (Clostridiaceae, Ruminococcus, etc.) decreased continuously. In contrast, the structure of gut microbiota was relatively stable during the intervention of EGCG on colon carcinogenesis. Enrichment of probiotics (Bifidobacterium, Lactobacillu, etc.) might be a potential mechanism for EGCG’s effects on tumor suppression. Via bioinformatics analysis, principal coordinate analysis and cluster analysis of the tumor formation process, we found that the diversity of gut microbiota increased in the tumor model group while that in the EGCG interfered group (EG) remained relatively stable.CONCLUSIONGut microbiota imbalance might be a potential mechanism for the prevention of malignant transformation by EGCG, which is significant for diagnosis, treatment, prognosis evaluation, and prevention of colorectal cancer.

Characterization of Inhibitor Binding Through Multiple Inhibitor Analysis: A Novel Local Fitting Method.

Understanding inhibitor binding modes is a key aspect of drug development. Early in a drug discovery effort these considerations often impact hit finding strategies and hit prioritization. Multiple inhibitor experiments, where enzyme inhibition is measured in the presence of two simultaneously varied inhibitors, can provide valuable information about inhibitor binding. These experiments utilize the inhibitor concentration dependence of the observed combined inhibition to determine the relationship between two compounds. In this way, it can be determined whether two inhibitors bind exclusively, independently, synergistically, or antagonistically. Novel inhibitors can be tested against each other or reference compounds to assist hit classification and characterization of inhibitor binding. In this chapter, we discuss the utility and design of multiple inhibitor experiments and present a new local curve fitting method for analyzing these data utilizing IC50 replots. The IC50 replot method is analogous to that used for determining mechanisms of inhibition with respect to substrate, as originally proposed by Cheng and Prusoff (Cheng and Prusoff Biochem Pharmacol 22: 3099-3108, 1973). The IC50 replot generated by this method reveals distinct patterns that are diagnostic of the nature of the interaction between two inhibitors. Multiple inhibition of the histone methyltransferase EZH2 by EPZ-5687 and the reaction product S-adenosylhomocysteine is presented as an example of the method.

Mo1673 Metagenomic Analysis of the Dynamic Changes in the Gut Microbiome of African American With Colorectal Cancer

Hassan Brim, Edward Lee, Scot Dowd, Adeyinka Laiyemo, Hassan Ashktorab Background: Linkage of specific bacterial markers to colorectal pathogenesis has been hampered by limited knowledge of the colonic microbiota and changing bacterial classification schemes. More than 80% of the colonic microbiota is not cultivatable and can only be assessed through metagenomic analysis. Aim: To perform a metagenomic analysis of 10 colon cancer tumors and their matched normal tissues. Methods: Metagenome sequencing was conducted on the illumina 2x150bp sequencing platform. Briefly DNA was prepared and libraries created using nextera library preparation kits. Sequencing was performed on each sample. For all samples an average of 20 million reads were obtained. Paired Sequencing reads were joined and entered into MG-RAST for metagenome analysis. The data was compared to M5NR using a maximum e-value of 1e-5, a minimum identity of 60 %, and a minimum alignment length of 15 measured in amino acids for protein databases. Results: Metagenomes were compared using best hit classification in two groups (normal and tumor). Based upon this best hit classification at the kingdom level in MG-RAST it was found, using one-way ANOVA, that tumor samples had a significantly higher classification related to Microviridae (bacteriophage) specifically related to Enterobacteria phage with a P= 0.0028. The other notable significant difference was the archaea. It was found at the class level that methanobacteria had higher classification hits in the normal samples (p = 0.03). Based upon functional gene abundance classification it was found that phage (P=0.007) as well as iron acquisition and metabolism genes (p= 0.05) were significantly different between the groups. Further analysis showed that as expected from the taxonomic classification data that tumor samples had higher levels of phage related functional classifications primarily related to phage capsid proteins. Other notable findings consisted of bacterial protein acetylation and deacetylation genes that were found to be significantly higher in normal tissue compared to tumor.

Mo1674 Upregulation of Mucin Production by a Probiotic-Derived Protein Through Activation of EGF Receptor

Hassan Brim, Edward Lee, Scot Dowd, Adeyinka Laiyemo, Hassan Ashktorab Background: Linkage of specific bacterial markers to colorectal pathogenesis has been hampered by limited knowledge of the colonic microbiota and changing bacterial classification schemes. More than 80% of the colonic microbiota is not cultivatable and can only be assessed through metagenomic analysis. Aim: To perform a metagenomic analysis of 10 colon cancer tumors and their matched normal tissues. Methods: Metagenome sequencing was conducted on the illumina 2x150bp sequencing platform. Briefly DNA was prepared and libraries created using nextera library preparation kits. Sequencing was performed on each sample. For all samples an average of 20 million reads were obtained. Paired Sequencing reads were joined and entered into MG-RAST for metagenome analysis. The data was compared to M5NR using a maximum e-value of 1e-5, a minimum identity of 60 %, and a minimum alignment length of 15 measured in amino acids for protein databases. Results: Metagenomes were compared using best hit classification in two groups (normal and tumor). Based upon this best hit classification at the kingdom level in MG-RAST it was found, using one-way ANOVA, that tumor samples had a significantly higher classification related to Microviridae (bacteriophage) specifically related to Enterobacteria phage with a P= 0.0028. The other notable significant difference was the archaea. It was found at the class level that methanobacteria had higher classification hits in the normal samples (p = 0.03). Based upon functional gene abundance classification it was found that phage (P=0.007) as well as iron acquisition and metabolism genes (p= 0.05) were significantly different between the groups. Further analysis showed that as expected from the taxonomic classification data that tumor samples had higher levels of phage related functional classifications primarily related to phage capsid proteins. Other notable findings consisted of bacterial protein acetylation and deacetylation genes that were found to be significantly higher in normal tissue compared to tumor.

Are similarity- or phylogeny-based methods more appropriate for classifying internal transcribed spacer (ITS) metagenomic amplicons?

• The internal transcribed spacer (ITS) of the nuclear ribosomal DNA region is a widely used species marker for plants and fungi. Recent metagenomic studies using next-generation sequencing, however, generate only partial ITS sequences. Here we compare the performance of partial and full-length ITS sequences with several classification methods. • We compiled a full-length ITS data set and created short fragments to simulate the read lengths commonly recovered from current next-generation sequencing platforms. We compared recovery, erroneous recovery, and coverage for the following methods: best BLAST hit classification, MEGAN classification, and automated phylogenetic assignment using the Statistical Assignment Program (SAP). • We found that summarizing results with more inclusive taxonomic ranks increased recovery and reduced erroneous recovery. The similarity-based methods BLAST and MEGAN performed consistently across most fragment lengths. Using a phylogeny-based method, SAP runs with queries 400 bp or longer worked best. Overall, BLAST had the highest recovery rates and MEGAN had the lowest erroneous recovery rates. • A high-throughput ITS classification method should be selected, taking into consideration read length, an acceptable tradeoff between maximizing the total number of classifications and minimizing the number of erroneous classifications, and the computational speed of the assignment method.

Online track processor for the CDF upgrade

A trigger track processor is being designed for the CDF upgrade. This processor identifies high momentum (P/sub T/>1.5 GeV/c) charged tracks in the new central outer tracking chamber for CDF II. The track processor is called the eXtremely Fast Tracker (XFT). The XFT design is highly parallel to handle the input fate of 183 Gbits/sec and output rate of 44 Gbits/sec. The processor is pipelined and reports the results for a new event every 132 ns. The processor uses three stages, hit classification, segment finding, and segment linking. The pattern recognition algorithms for the three stages are implemented in programmable logic devices (PLDs) which allow for in-situ modification of the algorithm at any time. The PLDs reside on three different types of modules. Prototypes of each of these modules have been designed and built, and are presently undergoing testing. An overview of the track processor and results of testing are presented.

System architecture and hardware design of the CDF XFT online track processor

A trigger track processor is being designed for CDF Run 2. This processor identifies high momentum (P/sub T/>1.5 GeV/c) charged tracks in the new central outer tracking chamber for the CDF II detector. The design of the track processor, called the eXtremely Fast Tracker (XFT), is highly parallel and handle an input rate of 183 Gbits/sec and output rate of 44 Gbits/sec. The XFT is pipelined and reports the results for a new event every 132 ns. The XFT uses three stages, hit classification, segment finding, and segment linking. The pattern recognition algorithms for the three stages are implemented in Programmable Logic Devices (PLDs) which allow for in-situ modification of the algorithm at any time. The PLDs reside on three different types of modules. Prototypes of each of these modules have been designed and built, and are working. An overview of the hardware design and the system architecture are presented.