B-acute lymphoblastic leukemia (B-ALL) consists of dozens of subtypes defined by distinct gene expression profiles (GEPs) and various genetic lesions. With the application of transcriptome sequencing (RNA-seq), multiple novel subtypes have been identified, which lead to an advanced B-ALL classification and risk-stratification system. However, the complexity of analyzing RNA-seq data for B-ALL classification hinders the implementation of the new B-ALL taxonomy. Here, we introduce MD-ALL (Molecular Diagnosis of ALL), an integrative platform featuring sensitive and accurate B-ALL classification based on GEPs and sentinel genetic alterations from RNA-seq data. In this study, we systematically analyzed 2,955 B-ALL RNA-seq samples and generated a reference dataset representing all the reported B-ALL subtypes. Using multiple machine learning algorithms, we identified the feature genes and then established highly sensitive and accurate models for B-ALL classification using either bulk or single-cell RNA-seq data. Importantly, this platform integrates multiple aspects of key genetic lesions acquired from RNAseq data, which include sequence mutations, large-scale copy number variations, and gene rearrangements, to perform comprehensive and definitive B-ALL classification. Through validation in a hold-out cohort of 974 samples, our models demonstrated superior performance for B-ALL classification compared with alternative tools. Moreover, to ensure accessibility and user-friendly navigation even for users with limited or no programming background, we developed an interactive graphical user interface for this MD-ALL platform, using the R Shiny package. In summary, MD-ALL is a user-friendly B-ALL classification platform designed to enable integrative, accurate, and comprehensive B-ALL subtype classification. MD-ALL is available from https://github.com/gu-lab20/MD-ALL.