Divergência genética entre linhagens parcialmente endogâmicas de milho pipoca por métodos multivariados e redes neurais artificiais

Abstract

In the selecting genitors process for diallel crosses, aiming to the production of hybrids, the evaluation of genetic divergence is widely used and can be measured by different methodologies. The multivariate analyzes is widely applied for the evaluation of genetic dissimilarity, highlighting the Tocher clustering method, unweighted mean linkage between groups (UPGMA) and canonical variables. Recently, the techniques of artificial neural networks have been used, using the map model of Kohonen aiming at the organization of genetic diversity. This work was carried out with the objective of estimating the genetic divergence among 20 partially endogenous popcorn strains by multivariate methods and artificial neural networks, aiming the identification of promising lines for diallel crosses. The experiments were conducted, on a field area, at Big Dourados Federal University experimental farm, in Dourados - Ms. In harvest I (agricultural year 2014/2015) and harvest II (agricultural year 2016/2017). Twenty popcorn S2 lineages (crop I) and S3 (crop II) were installed in the randomized block design, with three replications. The following characteristics were evaluated: days for female and male flowering (FF; FM), plant height (AP), and first ear insertion (AE), stem diameter (DC), prolificacy (PROLIF), broken plants (AQ), number of spikes autofecunded per plot (NEAP), ear length (CPE), spike diameter (DE), number of grains rows (NFG), hundred grain mass (M100), grain yield (PROD), grain expansion capacity (CE) and popcorn volume per hectare (PV). Multivariate statistical analyzes were performed for the relative importance of the characters, grouping methods of Tocher, UPGMA, more distant neighbor, canonical variables and the technique of artificial neural networks with the Kohonen map. Lineage choices were generally based on classifications of the methods and characters contribution. The most divergent and potential strains identified in S2 lineages were 9 (with lower AP, PROD and VP), 12 (with higher NFG and NEAP), 15 (with higher PROD, VP, CE and PROLIF) and 6 (with higher DC and lower FF and FM); and in the S3 lineages 2 (lowest CE and VP), 3 (highest CE), 10 (highest PROD, VP, M100 and NEAP) and 8 (higher DE, FF and FM). The characters that presented the greatest relative contribution to the divergence and differentiation of the lineages, for the S2 lineages were the PROD, VP, M100 and CE; and for the S3 lineages the characters CE, VP, AP and NEAP. Among the multivariate methods, the canonical variables and the UPGMA presented the greatest similarity among groups. Among the groupings of multivariate methods and artificial neural networks of Kohonen maps, the greatest relationship occurred among the results presented by canonical variables. The genetic distinct composition from the S2 and S3 lineages and the environmental differences of the evaluated agricultural years influenced the differences obtained in relation to the genetic divergence in the two experiments.