The paper presents an improved algorithm for gene identification in metagenomic sequences, which are derived from shotgun sequencing of microbial communities. The method, known as the "heuristic model," estimates parameters from the dependencies between oligonucleotide frequencies in protein-coding regions and genome nucleotide composition. The original version of this method was proposed in 1999 and has been used for reconstructing codon frequency vectors and initializing parameters for self-training gene-finding algorithms. With the advent of new prokaryotic genomes, the authors enhanced the method by using direct polynomial and logistic approximations of oligonucleotide frequencies and separating models for bacteria and archaea. This refinement has increased the accuracy of model reconstruction and gene prediction. The refined method is assessed on known prokaryotic genomes split into short sequences and shown to add thousands of new genes to existing annotations of human and mouse gut metagenomes. The study also highlights the importance of accurate gene prediction in metagenomics for understanding host metabolism and microbial communities.The paper presents an improved algorithm for gene identification in metagenomic sequences, which are derived from shotgun sequencing of microbial communities. The method, known as the "heuristic model," estimates parameters from the dependencies between oligonucleotide frequencies in protein-coding regions and genome nucleotide composition. The original version of this method was proposed in 1999 and has been used for reconstructing codon frequency vectors and initializing parameters for self-training gene-finding algorithms. With the advent of new prokaryotic genomes, the authors enhanced the method by using direct polynomial and logistic approximations of oligonucleotide frequencies and separating models for bacteria and archaea. This refinement has increased the accuracy of model reconstruction and gene prediction. The refined method is assessed on known prokaryotic genomes split into short sequences and shown to add thousands of new genes to existing annotations of human and mouse gut metagenomes. The study also highlights the importance of accurate gene prediction in metagenomics for understanding host metabolism and microbial communities.