The program follows a multi-step process to obtain the desired patterns. First, it performs a pattern search by browsing the file. It uses a regular expression to find patterns of the specified length, ending with the pattern chosen by the user.

After obtaining the gene patterns, the program repeats this operation for the reference genome if the user has supplied it. Next, it performs a binary search to retain only those motifs that are exclusively present in the gene database and not in the genome database.

Once the motifs have been extracted, if the user has selected the mutation option, the program performs mutations for each gene-pattern pair to obtain the desired nucleotide sequence. The mutations are applied directly to the nucleotide sequence. Once the mutations have been made, the sequence is translated into proteins.

The program then counts the number of stop codons to determine whether the mutation has resulted in the appearance of a stop codon. In addition, it assesses for each motif whether the mutation will inevitably result in a stop codon or whether certain mutations will result in a different codon. All this information is stored in a file in CSV format.

We then search for the minimum number of guide RNAs capable of binding to many of our class genes. To do so, we transform our list of genes into an array of 0 and 1, where each index represents a gene. The association between a gene and its index is stored in a dictionary for easy access. We associate a gene array for each guide RNA motif where every gene it binds is represented by a 1 at their index and a 0 when it doesn't. Now that our motifs are associated with their gene array, we filter them by checking if a gene array includes another. To do so, we go through our list of motifs sorted by the number of genes they cover, starting with the ones with the most coverage.

In this loop, for each motif, we go through our list a second time, but this time, we begin with the motifs that cover the least amount of genes. We then compare their gene array. If all the genes of the least covering motifs are included in the most covering motifs, the least covering motif is filtered out and will not be processed in the first loop. With this method, we can discard motifs with redundant covering. This method filters out a lot of motifs, but there is still some overlap. Because even if a motif gene array is not included in another, the combination of other gene arrays can include it. That is why we compute the importance of each motif.

Firstly, we go through all the gene arrays and add them up. We get a full array where each index is at least one. Overlapping is represented by cells with a number superior to 1. With this full array, we remove each motif from the full array and check for each motif how many cells in the full gene array became zeros. It gives us the impact of a motif in a gene. If a motif has the same impact as its number of genes covered, it is essential, as no combination of motifs can cover its genes. Then, we sort our motifs by their impact. We then search for the best combination of genes thanks to the impact. At the end, we obtain a list ordered by gene importance. We call importance the value that a motif adds to the overall coverage. Therefore, the first motif in this list is optimal if we want to use only one guide RNA. The second one is optimal if used with the first one to get the maximum coverage with two guide RNAs, and so on. The files output for this part are two CSV files. One contains the sorted list of motifs by importance with the percentage of coverage if combined with the previous ones. The second is the list of motifs with their associated genes ordered by their importance.

However, it is important to note that the training data was based on the E. Coli genome, which differs considerably from the multiple species present in our beta-lactamase database.