Objective To construct core collections of bitter gourd (Momordica charantia) which can represent the genetic diversity of the initial population by comparing different construction methods, and facilitate the efficient utilization of bitter gourd germplasm.

Method We sampled 154 bitter gourd germplasm as materials, predicted the genotypic values of five traits (node bearing first female flower, fruit length, fruit width, flesh thickness and mass per fruit) without bias using mixed linear model analysis, calculated the genetic distances among bitter gourd germplasm based on genotypic values of all five traits using Mahalanobis distance, and constructed core collections with 30% sampling proportion by using eight different clustering methods and three different sampling strategies. The quality of core collections constructed using different clustering methods and sampling strategies were evaluated.

Result The variation coefficients of all five traits of the core collections constructed with eight clustering methods were larger than those of the initial population. Single linkage was better compared to the other seven clustering methods by significantly increasing the variances and coefficients of variation for all five traits. The ranges of core collections constructed with preferred sampling and deviation sampling were consistent with those of the initial population. The variation coefficients of three traits of the core collection from deviation sampling were larger compared to the other two sampling methods, indicating that deviation sampling was slightly better than random sampling and preferred sampling. Forty six core collections of bitter gourd were obtained based on Mahalanobis distance, deviation sampling and single linkage. Among them, Y5, Y87, Y112 and Y139 were the backbone materials.

Conclusion The 46 core collections of bitter gourd, which were obtained based on Mahalanobis distance, deviation sampling and single linkage, can well represent the genetic diversity of the initial population. This study further provides evidence for high genetic diversity in the bitter gourd germplasm from India and Southeast Asian. Our results provide an important theoretical basis for the collection, evaluation and efficient utilization of bitter gourd genetic resources.