This section of the article discusses the new features and improvements in RAxML-NG compared to its predecessors, RAxML and ExaML. Key enhancements include: 1. **Evolutionary Model Extensions**: - **New DNA Models**: Full support for all 22 classical GTR-derived models, allowing flexible parametrization and partitioning. - **Flexible Models for Multi-State Data**: Support for up to 64 states and user-defined encoding for multi-state sequence data, including ambiguous characters. 2. **Rate Heterogeneity**: - **FreeRate Model**: Support for a flexible rate heterogeneity model that does not rely on a priori rate distribution. - **Branch Length Linkage**: Support for three types of branch linkage models (linked, unlinked, scaled) to better model evolutionary rates across partitions. 3. **Search Algorithm Modifications**: - **Subtree Enumeration**: Improved subtree enumeration procedure to ensure promising moves are not skipped. - **Transfer Bootstrap**: Optimized computation of Transfer Bootstrap Expectation (TBE) support values, which are faster and require less memory. 4. **Performance and Scalability**: - **Fine-Grained Parallelization**: Efficient parallelization using MPI and pthreads, with load balancing and checkpointing capabilities. - **Scalability**: Linear scalability up to 1,024 cores on large partitioned multi-gene alignments, with superlinear speedups on DNA datasets. 5. **Evaluation**: - **Experimental Setup**: Benchmarking runs on a cluster with 224 compute nodes, comparing RAxML-NG to IQTree, RAxML, and ExaML. - **Results**: RAxML-NG shows superior performance in terms of search efficiency, inference times, and scalability, particularly on taxon-rich datasets. These improvements make RAxML-NG a more powerful and user-friendly tool for maximum likelihood phylogenetic inference, especially for large and complex datasets.This section of the article discusses the new features and improvements in RAxML-NG compared to its predecessors, RAxML and ExaML. Key enhancements include: 1. **Evolutionary Model Extensions**: - **New DNA Models**: Full support for all 22 classical GTR-derived models, allowing flexible parametrization and partitioning. - **Flexible Models for Multi-State Data**: Support for up to 64 states and user-defined encoding for multi-state sequence data, including ambiguous characters. 2. **Rate Heterogeneity**: - **FreeRate Model**: Support for a flexible rate heterogeneity model that does not rely on a priori rate distribution. - **Branch Length Linkage**: Support for three types of branch linkage models (linked, unlinked, scaled) to better model evolutionary rates across partitions. 3. **Search Algorithm Modifications**: - **Subtree Enumeration**: Improved subtree enumeration procedure to ensure promising moves are not skipped. - **Transfer Bootstrap**: Optimized computation of Transfer Bootstrap Expectation (TBE) support values, which are faster and require less memory. 4. **Performance and Scalability**: - **Fine-Grained Parallelization**: Efficient parallelization using MPI and pthreads, with load balancing and checkpointing capabilities. - **Scalability**: Linear scalability up to 1,024 cores on large partitioned multi-gene alignments, with superlinear speedups on DNA datasets. 5. **Evaluation**: - **Experimental Setup**: Benchmarking runs on a cluster with 224 compute nodes, comparing RAxML-NG to IQTree, RAxML, and ExaML. - **Results**: RAxML-NG shows superior performance in terms of search efficiency, inference times, and scalability, particularly on taxon-rich datasets. These improvements make RAxML-NG a more powerful and user-friendly tool for maximum likelihood phylogenetic inference, especially for large and complex datasets.