Investigation of muscle transcriptomes using gradient boosting machine learning identifies molecular predictors of feed efficiency in growing pigs

Abstract

Background Improving feed efficiency (FE) is a major challenge in pig production. This complex trait is characterized by a high variability. Therefore, the identification of predictors of FE may be a relevant strategy to reduce phenotyping efforts in breeding and selection programs. The aim of this study was to investigate the suitability of expressed muscle genes in prediction of FE traits in growing pigs. The approach considered different transcriptomics experiments to cover a large range of FE values and identify reliable predictors. Results Microarrays data were obtained from longissimus muscles of two lines divergently selected for residual feed intake (RFI). Pigs ( n = 71) from three experiments belonged to generations 6 to 8 of selection, were fed either a diet with a standard composition or a diet rich in fiber and lipids, received feed ad libitum or at restricted level, and weighed between 80 and 115 kg at slaughter. For each pig, breeding value for RFI was estimated (RFI-BV), and feed conversion ratio (FCR) and energy-based feed conversion ratio (FCRe) were calculated during the test periods. Gradient boosting algorithms were used on the merged muscle transcriptomes to identify very important predictors of FE traits. About 20,405 annotated molecular probes were commonly expressed in longissimus muscle across experiments. Six to 267 expressed muscle genes covering a variety of biological processes were found as important predictors for RFI-BV ( R 2 = 0.63–0.65), FCR ( R 2 = 0.61–0.70) and FCRe ( R 2 = 0.49–0.52). The error of prediction was less than 8% for FCR. Altogether, 56 predictors were common to RFI-BV and FCR. Expression levels of 24 target genes were further measured by qPCR. Linear regression confirmed the good accuracy of combining mRNA levels of these genes to fit FE traits (RFI-BV: R 2 = 0.73, FRC: R 2 = 0.76; FCRe: R 2 = 0.75). Stepwise regression procedure highlighted 10 genes (FKBP5, MUM1, AKAP12, FYN, TMED3, PHKB, TGF, SOCS6, ILR4, and FRAS1) in a linear combination predicting FCR and FCRe. In addition, FKBP5 and expression levels of five other genes (IGF2, SERINC3, CSRNP3, EZR and RPL16) significantly contributed to RFI-BV. Conclusion It was possible to identify few genes expressed in muscle that might be reliable predictors of feed efficiency. Electronic supplementary material The online version of this article (10.1186/s12864-019-6010-9) contains supplementary material, which is available to authorized users.

Identification of Predictor Genes for Feed Efficiency in Beef Cattle by Applying Machine Learning Methods to Multi-Tissue Transcriptome Data

“…Therefore, accurately predicting FE and related traits from molecular datasets is not straightforward. So far, several studies have explored the feasibility of identifying molecular predictors for FE using different ML algorithms [e.g., Clemmons et al, 2019 (Beef cattle), Messad et al, 2019 (pigs), and Piles et al, 2019 (pigs)]. However, none of these attempted to compare the prediction performance of combining two ML methods together.…”

“…To date, several studies have reported the application of different ML methods in prediction for FE. For example, Messad et al (2019) successfully tested the reliability of gradient tree boosting (XGBoost) in identifying molecular predictors of FE in pigs; Yao et al (2013) found that Random Forest (RF) could be used effectively to identify additive predictors associated with FE in cattle; and support vector machine (SVM) had also been proven to be a reliable method in genomic prediction of FE in dairy cattle ( Yao et al, 2016 ). Piles et al (2019) found that out of four ML methods used [RF, SVM, Elastic Net (ENET), and nearest shrunken centroid], ENET produced the best classification accuracy of residual feed intake (RFI) in pigs using 200 selected genes from liver.…”

Genome Wide Assessment of Genetic Variation and Population Distinctiveness of the Pig Family in South Africa

“…The number of mature individuals is estimated to be between 2000 and 5000 in the Kruger National Park (Ferreira et al, 2013). The geographical separation of the three national parks from which the warthogs were sampled, could have created small and fragmented subpopulations leading to escalated F IS values (Lamsyah et al, 2009) rs80854994 4 106,719,032 BCL2L15 Mastitis (Chen et al, 2015) Villages and DUR rs81282695 6 94,442,844 POU3F1 Neurobehavioral functioning (Eusebi et al, 2018) rs81282695 6 94442844 FHL3 Carcass traits (Zuo et al, 2004(Zuo et al, , 2007 Villages and KOL rs81430450 11 24,063,007 DNAJC15 Feeding efficiency (Reyer et al, 2017a) rs81430450 11 24,063,007 EPSTI1 Fertility traits (Gaddis et al, 2016), fat deposition (Zhang et al, 2018) SAL&LWT (Luo et al, 2012); carcass weight (Kang et al, 2013) Villages and WBO rs81244815 2 50,167,007 SWAP70 Disease resistance (Ma et al, 2011;Zhang et al, 2018) rs81244815 2 50,167,007 SBF2 Fertility (Zhang et al, 2014); immune function (Ibeagha-Awemu et al, 2016) rs81401075 8 73,841,435 FRAS1 Sow reproductive traits (Fischer et al, 2015), feed efficiency (Messad et al, 2019) rs81401075 8 73,841,435 NPY2R Obesity (Siddiq et al, 2007;Hunt et al, 2011) Villages (Smith et al, 2019); body width in gilts and sows (Rothschild, 2010), body weight traits (Borowska et al, 2017), altitude (Zhang et al, 2014) rs81478390 13 53,707,241 RYBP Body conformation traits -body weight, body length, body height, and chest circumference (Zhou et al, 2016) rs81330369 9 7,449,894 FCHSD2 Milk production traits (Kemper et al, 2015) rs80975991 7 33,481,446 ZFAND3 Growth and carcass quality traits (Li and Kim, 2015) rs80855522 4 11,0552,282 GNAI3 Heat tolerance (Berihulay et al, 2019) rs80988392 1 213,780,848 PTPRD Meat quality (Raschetti et al, 2013) due to Wahlund effect. As expected, we found that the village pig populations of South Africa had high inbreeding values compared with other popul...…”

The CSRNP Gene Family Serves as a Prognostic Biomarker in Clear Cell Renal Cell Carcinoma

“…CSRNP3 was found to encode a transcriptional factor for muscle development in growing pigs ( 25 ), and was reported as a target gene to treat obesity and metabolic syndrome in an exome-wide mediated study ( 26 ). However, the role of CSRNP3 in cancer development requires further investigation.…”