The paper presents a new framework for genotype imputation that leverages a cosmopolitan reference panel and a computational approximation to improve accuracy and efficiency. The authors develop an algorithm that uses local sequence similarity to select a custom reference panel for each study haplotype in each genomic region, reducing the computational burden while maintaining accuracy. This approach is evaluated using data from HapMap 3 and the MalariaGEN Project, demonstrating high accuracy in various human populations, especially at low-frequency variants. The framework is shown to be faster and more accurate than existing methods like Beagle when using large, sequence-based reference panels. The authors also discuss general computational strategies for balancing efficiency and accuracy and provide practical recommendations for imputation-based studies. The methodology is implemented in the IMPUTE2 software package.The paper presents a new framework for genotype imputation that leverages a cosmopolitan reference panel and a computational approximation to improve accuracy and efficiency. The authors develop an algorithm that uses local sequence similarity to select a custom reference panel for each study haplotype in each genomic region, reducing the computational burden while maintaining accuracy. This approach is evaluated using data from HapMap 3 and the MalariaGEN Project, demonstrating high accuracy in various human populations, especially at low-frequency variants. The framework is shown to be faster and more accurate than existing methods like Beagle when using large, sequence-based reference panels. The authors also discuss general computational strategies for balancing efficiency and accuracy and provide practical recommendations for imputation-based studies. The methodology is implemented in the IMPUTE2 software package.