gppf

gppf reconstructs evolutionary trees from mutation frequency data by allowing mutation losses and recurrent acquisitions to relax the infinite sites assumption, enabling more accurate clonal reconstruction in cancer genomics.

Key Features:

• Relaxation of Infinite Sites Assumption: Allows loss of previously acquired mutations and multiple acquisitions (recurrent mutations) to model evolutionary scenarios beyond the Perfect Phylogeny model.
• ILP-Based Formulation: Formulates evolutionary tree reconstruction as an Integer Linear Programming (ILP) problem to represent and solve complex evolutionary constraints.
• Clonal Reconstruction from Mutation Frequencies: Uses input data representing the fraction of cells with specific mutations across samples to infer clonal structures and mutation histories.
• Model Flexibility: Supports multiple evolutionary models including perfect, persistent, dollo, and caminsokal to accommodate different mutational behaviors.
• Problem Scope: Addresses the Incomplete Directed Phylogeny problem and the Clonal Reconstruction problem.
• Experimental Analysis and Comparative Advantage: Evaluated on real and simulated datasets and demonstrates improved interpretation over traditional Perfect Phylogeny models when mutation losses are present.

Scientific Applications:

• Clonal Reconstruction in Cancer Genomics: Infers tumor evolutionary histories and clonal composition from mutation-frequency data across cancer samples.
• Inference of Recurrent and Back Mutations: Enables detection and modeling of recurrent and back mutations in tumor evolution by relaxing the infinite sites assumption.
• Comparative Model Evaluation: Allows testing and selection among perfect, persistent, dollo, and caminsokal models for specific datasets.
• Benchmarking and Validation: Supports evaluation on real and simulated datasets to assess model fit and the impact of mutation losses.

Methodology:

gppf formulates evolutionary tree reconstruction as an Integer Linear Programming (ILP) problem on input matrices of mutation frequencies (fractions of cells with specific mutations across samples), incorporates constraints to allow mutation loss and multiple acquisitions, supports perfect, persistent, dollo, and caminsokal models, and was evaluated on real and simulated datasets.