Predicting and improving complex beer flavor through machine learning.

The perception and appreciation of food flavor depends on many interacting chemical compounds and external factors, and therefore proves challenging to understand and predict. Here, we combine extensive chemical and sensory analyses of 250 different beers to train machine learning models that allow predicting flavor and consumer appreciation. For each beer, we measure over 200 chemical properties, perform quantitative descriptive sensory analysis with a trained tasting panel and map data from over 180,000 consumer reviews to train 10 different machine learning models. The best-performing algorithm, Gradient Boosting, yields models that significantly outperform predictions based on conventional statistics and accurately predict complex food features and consumer appreciation from chemical profiles. Model dissection allows identifying specific and unexpected compounds as drivers of beer flavor and appreciation. Adding these compounds results in variants of commercial alcoholic and non-alcoholic beers with improved consumer appreciation. Together, our study reveals how big data and machine learning uncover complex links between food chemistry, flavor and consumer perception, and lays the foundation to develop novel, tailored foods with superior flavors.

Massive QTL analysis identifies pleiotropic genetic determinants for stress resistance, aroma formation, and ethanol, glycerol and isobutanol production in Saccharomyces cerevisiae.

BACKGROUND: The brewer's yeast Saccharomyces cerevisiae is exploited in several industrial processes, ranging from food and beverage fermentation to the production of biofuels, pharmaceuticals and complex chemicals. The large genetic and phenotypic diversity within this species offers a formidable natural resource to obtain superior strains, hybrids, and variants. However, most industrially relevant traits in S. cerevisiae strains are controlled by multiple genetic loci. Over the past years, several studies have identified some of these QTLs. However, because these studies only focus on a limited set of traits and often use different techniques and starting strains, a global view of industrially relevant QTLs is still missing. RESULTS: Here, we combined the power of 1125 fully sequenced inbred segregants with high-throughput phenotyping methods to identify as many as 678 QTLs across 18 different traits relevant to industrial fermentation processes, including production of ethanol, glycerol, isobutanol, acetic acid, sulfur dioxide, flavor-active esters, as well as resistance to ethanol, acetic acid, sulfite and high osmolarity. We identified and confirmed several variants that are associated with multiple different traits, indicating that many QTLs are pleiotropic. Moreover, we show that both rare and common variants, as well as variants located in coding and non-coding regions all contribute to the phenotypic variation. CONCLUSIONS: Our findings represent an important step in our understanding of the genetic underpinnings of industrially relevant yeast traits and open new routes to study complex genetics and genetic interactions as well as to engineer novel, superior industrial yeasts. Moreover, the major role of rare variants suggests that there is a plethora of different combinations of mutations that can be explored in genome editing.

Mapping the genomic diversity of HCV subtypes 1a and 1b: Implications of structural and immunological constraints for vaccine and drug development.

Despite significant progress in hepatitis C (HCV) treatment, global viral eradication remains a challenge. An in-depth map of its genome diversity within the context of structural and immunological constraints could contribute to the design of pan-genotypic antivirals and preventive vaccines. For such analyses, extensive information is only available for the highly prevalent HCV genotypes (GT) 1a and 1b. Using 647 GT1a and 408 GT1b full-genome sequences obtained from the Los Alamos database, we found that respectively 3 per cent and 82 per cent of all codon positions are under positive and negative selective pressure, suggesting variation mainly accumulates due to random genetic drift. An association between conservation and both structured RNA and secondary protein structures confirmed the important role of structural elements at nucleotide and at amino acid level. Remarkably, CD8+ T-cell epitopes in HCV GT1a were significantly more conserved, while at the same time containing more sites under positive selection. Similarly, CD4+ T-cell epitopes were significantly more conserved in both HCV subtypes, but under less positive selective pressure in GT1b and more negative selective pressure in GT1a. In contrast, B-cell epitopes in both subtypes were less conserved and under less stringent negative selection. These findings argue against immune selective pressure as the main force of between-host diversifying evolution. Despite its high variability, HCV is under strict evolutionary constraints, most probably to keep its genes and proteins functional during the replication cycle. These are encouraging findings for vaccine and drug design, which could consider these newly established genetic diversity profiles.

Genetic Diversity and Selective Pressure in Hepatitis C Virus Genotypes 1-6: Significance for Direct-Acting Antiviral Treatment and Drug Resistance.

Treatment with pan-genotypic direct-acting antivirals, targeting different viral proteins, is the best option for clearing hepatitis C virus (HCV) infection in chronically infected patients. However, the diversity of the HCV genome is a major obstacle for the development of antiviral drugs, vaccines, and genotyping assays. In this large-scale analysis, genome-wide diversity and selective pressure was mapped, focusing on positions important for treatment, drug resistance, and resistance testing. A dataset of 1415 full-genome sequences, including genotypes 1-6 from the Los Alamos database, was analyzed. In 44% of all full-genome positions, the consensus amino acid was different for at least one genotype. Focusing on positions sharing the same consensus amino acid in all genotypes revealed that only 15% was defined as pan-genotypic highly conserved (≥99% amino acid identity) and an additional 24% as pan-genotypic conserved (≥95%). Despite its large genetic diversity, across all genotypes, codon positions were rarely identified to be positively selected (0.23%-0.46%) and predominantly found to be under negative selective pressure, suggesting mainly neutral evolution. For NS3, NS5A, and NS5B, respectively, 40% (6/15), 33% (3/9), and 14% (2/14) of the resistance-related positions harbored as consensus the amino acid variant related to resistance, potentially impeding treatment. For example, the NS3 variant 80K, conferring resistance to simeprevir used for treatment of HCV1 infected patients, was present in 39.3% of the HCV1a strains and 0.25% of HCV1b strains. Both NS5A variants 28M and 30S, known to be associated with resistance to the pan-genotypic drug daclatasvir, were found in a significant proportion of HCV4 strains (10.7%). NS5B variant 556G, known to confer resistance to non-nucleoside inhibitor dasabuvir, was observed in 8.4% of the HCV1b strains. Given the large HCV genetic diversity, sequencing efforts for resistance testing purposes may need to be genotype-specific or geographically tailored.

An integrated map of HIV genome-wide variation from a population perspective.

BACKGROUND: The HIV pandemic is characterized by extensive genetic variability, which has challenged the development of HIV drugs and vaccines. Although HIV genomes have been classified into different types, groups, subtypes and recombinants, a comprehensive study that maps HIV genome-wide diversity at the population level is still lacking to date. This study aims to characterize HIV genomic diversity in large-scale sequence populations, and to identify driving factors that shape HIV genome diversity. RESULTS: A total of 2996 full-length genomic sequences from 1705 patients infected with 16 major HIV groups, subtypes and circulating recombinant forms (CRFs) were analyzed along with structural, immunological and peptide inhibitor information. Average nucleotide diversity of HIV genomes was almost 50% between HIV-1 and HIV-2 types, 37.5% between HIV-1 groups, 14.7% between HIV-1 subtypes, 8.2% within individual HIV-1 subtypes and less than 1% within single patients. Along the HIV genome, diversity patterns and compositions of nucleotides and amino acids were highly similar across different groups, subtypes and CRFs. Current HIV-derived peptide inhibitors were predominantly derived from conserved, solvent accessible and intrinsically ordered structures in the HIV-1 subtype B genome. We identified these conserved regions in Capsid, Nucleocapsid, Protease, Integrase, Reverse transcriptase, Vpr and the GP41 N terminus as potential drug targets. In the analysis of factors that impact HIV-1 genomic diversity, we focused on protein multimerization, immunological constraints and HIV-human protein interactions. We found that amino acid diversity in monomeric proteins was higher than in multimeric proteins, and diversified positions were preferably located within human CD4 T cell and antibody epitopes. Moreover, intrinsic disorder regions in HIV-1 proteins coincided with high levels of amino acid diversity, facilitating a large number of interactions between HIV-1 and human proteins. CONCLUSIONS: This first large-scale analysis provided a detailed mapping of HIV genomic diversity and highlighted drug-target regions conserved across different groups, subtypes and CRFs. Our findings suggest that, in addition to the impact of protein multimerization and immune selective pressure on HIV-1 diversity, HIV-human protein interactions are facilitated by high variability within intrinsically disordered structures.

A new ensemble coevolution system for detecting HIV-1 protein coevolution.

BACKGROUND: A key challenge in the field of HIV-1 protein evolution is the identification of coevolving amino acids at the molecular level. In the past decades, many sequence-based methods have been designed to detect position-specific coevolution within and between different proteins. However, an ensemble coevolution system that integrates different methods to improve the detection of HIV-1 protein coevolution has not been developed. RESULTS: We integrated 27 sequence-based prediction methods published between 2004 and 2013 into an ensemble coevolution system. This system allowed combinations of different sequence-based methods for coevolution predictions. Using HIV-1 protein structures and experimental data, we evaluated the performance of individual and combined sequence-based methods in the prediction of HIV-1 intra- and inter-protein coevolution. We showed that sequence-based methods clustered according to their methodology, and a combination of four methods outperformed any of the 27 individual methods. This four-method combination estimated that HIV-1 intra-protein coevolving positions were mainly located in functional domains and physically contacted with each other in the protein tertiary structures. In the analysis of HIV-1 inter-protein coevolving positions between Gag and protease, protease drug resistance positions near the active site mostly coevolved with Gag cleavage positions (V128, S373-T375, A431, F448-P453) and Gag C-terminal positions (S489-Q500) under selective pressure of protease inhibitors. CONCLUSIONS: This study presents a new ensemble coevolution system which detects position-specific coevolution using combinations of 27 different sequence-based methods. Our findings highlight key coevolving residues within HIV-1 structural proteins and between Gag and protease, shedding light on HIV-1 intra- and inter-protein coevolution.

HIV-1 Gag C-terminal amino acid substitutions emerging under selective pressure of protease inhibitors in patient populations infected with different HIV-1 subtypes.

HIV-1 Gag amino acid substitutions associated with protease inhibitor (PI) treatment have mainly been reported in subtype B, while information on other subtypes is scarce. Using sequences from 11613 patients infected with different HIV-1 subtypes, we evaluated the prevalence of 93 Gag amino acid substitutions and their association with genotypic PI resistance. A significant association was found for 13 Gag substitutions, including A431V in both subtype B and CRF01_AE. K415R in subtype C and S451G in subtype B were newly identified. Most PI-associated Gag substitutions are located in the flexible C-terminal domain, revealing the key role this region plays in PI resistance.