Identifying biosynthetic gene clusters
A BGC prediction system identifies candidate biosynthetic gene clusters (BGCs) within genomes using machine-learned models, such as a shallow neural network and recurrent neural network (RNN). A set of domains within a genome sequence are identified, each domain corresponds to a set of domain identifiers. A shallow neural network block is applied to each set of domain identifiers to produce a set of vectors. An RNN block is applied to the set of vectors to produce a BGC class score for each domain. The RNN block was trained using an identified set of positive vectors, which represents known BGCs, and a synthesized set of negative vectors, which is unlikely to represent BGCs. Candidate BGCs are selected by averaging BGC class scores across genes within a domain and comparing the average BGC class scores to a threshold. The candidate BGCs are provided for display on a user interface.
1. A method comprising:
identifying, in a genome sequence, a set of domains, each identified domain corresponding to a set of domain identifiers;
applying a shallow neural network block to each set of domain identifiers to produce a set of vectors, each vector corresponding to a set of domain identifiers;
applying a recurrent neural network (RNN) block to the set of vectors to produce a biosynthetic gene cluster (BGC) class score for each domain, wherein the RNN block was trained by:
identifying a set of positive vectors representing known BGCs;
synthesizing a set of negative vectors unlikely to represent BGCs;
applying the RNN block to the positive and negative sets of vectors to generate predictions of whether each vector is a positive or negative vector; and
updating weights of the RNN block based on the predictions;
selecting candidate BGCs by averaging BGC class scores across genes within a domain and comparing the average BGC class scores to a threshold;
predicting a molecular activity of biosynthetic products derived from the selected BGCs; and
providing for display, on a user interface, the candidate BGCs and predicted molecular activity.
2. The method of claim 1 , further comprising:
processing candidate BGCs, wherein processing includes merging and filtering candidate BGCs based on at least one of: a presence of known BGCs, a cluster length, or a distance between candidate BGCs.
3. The method of claim 1 , further comprising:
merging consecutive candidate BGC genes that are adjacent in the genome sequence.
4. The method of claim 1 , wherein the RNN block is a bi-directional long short-term memory (LSTM) block.
5. The method of claim 1 , wherein the domain identifiers are maintained in genomic order.
6. The method of claim 1 , wherein each vector in the set of vectors comprises one hundred elements, each element being a real number, each element representing a property of the domain based on its genomic context.
7. The method of claim 1 , further comprising:
predicting, for each candidate BGC, with a classifier, a secondary metabolite class based on a biosynthetic product and molecular activity of the candidate BGC.
8. The method of claim 7 , wherein the classifier is a random forest classifier.
9. The method of claim 1 , wherein the set of negative vectors are synthesized by:
retrieving a genome sequence with known BGCs;
modifying the genome sequence by replacing a portion of the genes within the known BGCs with random genes of similar length;
generating a set of identifiers for each domain in the modified genome sequence; and
applying a shallow neural network block to each domain in the modified genome sequence to produce a negative set of vectors.
10. The method of claim 1 , wherein applying the RNN block to the set of vectors further comprises applying a sigmoid activation function.
11. A non-transitory computer-readable storage medium containing computer program code comprising instructions that, when executed by a processor, causes the processor to:
identify, in a genome sequence, a set of domains, each identified domain corresponding to a set of domain identifiers;
apply a shallow neural network block to each set of domain identifiers to produce a set of vectors, each vector corresponding to a set of domain identifiers;
apply a recurrent neural network (RNN) block to the set of vectors to produce a biosynthetic gene cluster (BGC) class score for each domain, wherein the RNN block was trained by:
identifying a set of positive vectors representing known BGCs;
synthesizing a set of negative vectors unlikely to represent BGCs;
applying the RNN block to the positive and negative sets of vectors to generate predictions of whether each vector is a positive or negative vector; and
updating weights of the RNN block based on the predictions;
select candidate BGCs by averaging BGC class scores across genes within a domain and comparing the average BGC class scores to a threshold;
predict a molecular activity of biosynthetic products derived from the selected BGCs; and
provide for display, on a user interface, the candidate BGCs and predicted molecular activity.
12. The non-transitory computer-readable storage medium of claim 11 , wherein the instructions, when executed by the processor, further cause the processor to:
process candidate BGCs by merging and filtering candidate BGCs based on at least one of: a presence of known BGCs, a cluster length, or a distance between candidate BGCs.
13. The non-transitory computer-readable storage medium of claim 11 , wherein the instructions, when executed by the processor, cause the processor to:
merge consecutive candidate BGC genes.
14. The non-transitory computer-readable storage medium of claim 11 , wherein the RNN block is a bi-directional long short-term memory (LSTM) block.
15. The non-transitory computer-readable storage medium of claim 11 , wherein the domain identifiers are maintained in genomic order.
16. The non-transitory computer-readable storage medium of claim 11 , wherein each vector in the set of vectors comprises one hundred elements, each element being a real number, each element representing a property of the domain based on its genomic context.
17. The non-transitory computer-readable storage medium of claim 11 , wherein the instructions, when executed by the processor, further cause the processor to:
predict, for each candidate BGC, with a classifier, a secondary metabolite class based on a biosynthetic product and molecular activity of the candidate BGC.
18. The non-transitory computer-readable storage medium of claim 17 , wherein the classifier is a random forest classifier.
19. The non-transitory computer-readable storage medium of claim 17 , wherein the set of negative vectors are synthesized by:
retrieving a genome sequence with known BGCs;
modifying the genome sequence by replacing a portion of the genes within the known BGCs with random genes of similar length;
generating a set of identifiers for each domain in the modified genome sequence; and
applying a shallow neural network block to each domain in the modified genome sequence to produce a negative set of vectors.
20. The non-transitory computer-readable storage medium of claim 11 , wherein the instructions that cause the processor to apply the RNN block to the set of vectors further comprise instructions that cause the processor to apply a sigmoid activation function.