A deep learning genome-mining strategy for biosynthetic gene cluster prediction

Abstract

Natural products represent a rich reservoir of small molecule drug candidates utilized as antimicrobial drugs, anticancer therapies, and immunomodulatory agents. These molecules are microbial secondary metabolites synthesized by co-localized genes termed Biosynthetic Gene Clusters (BGCs). The increase in full microbial genomes and similar resources has led to development of BGC prediction algorithms, although their precision and ability to identify novel BGC classes could be improved. Here we present a deep learning strategy (DeepBGC) that offers reduced false positive rates in BGC identification and an improved ability to extrapolate and identify novel BGC classes compared to existing machine-learning tools. We supplemented this with random forest classifiers that accurately predicted BGC product classes and potential chemical activity. Application of DeepBGC to bacterial genomes uncovered previously undetectable putative BGCs that may code for natural products with novel biologic activities. The improved accuracy and classification ability of DeepBGC represents a major addition to in-silico BGC identification.

Supporting supervised learning in fungal Biosynthetic Gene Cluster discovery: new benchmark datasets

“…Supervised learning was shown to perform well at BGC discovery in previous work that focused on handling bacteria data [5], [6]. Given that annotated data are needed to perform a supervised learning approach, we propose here fungal BGC datasets to support the development of this approach for fungi.…”

“…To generate classification models based on a supervised learning method, we extracted Pfam [22] 6 IDs from the positive and negative instances. All datasets were converted into pfamtsv format [6], which is required as input in the supervised learning approach applied in this work. For each dataset, 80% were randomly selected for the training phase, while 20% were held out for the validation phase, as shown in Table I.…”

“…In this section we describe the methods applied to analyse the performance of a supervised learning approach using the fungal BGC datasets presented in Section III-A and the test data presented in Section III-B. To generate classification models with our fungal BGC datases, we utilized the Deep-BGC system [6]. DeepBGC executable, source code and other resources are openly available 8 .…”

“…Among these resources, there are pre-built BGC classification models and word2vec-based embeddings built using Pfam IDs, referred to as pfam2vec embeddings. In [6] the authors explained that pfam2vec embeddings were trained based in a skipgram architecture with 100 dimensions and over 15,686 unique Pfam IDs. DeepBGC classification is based on a Bidirectional Long Short Term Memory (BiLSTM) neural network, for which the input are pfam2vec embeddings.…”

“…Supervised learning has been previously used to predicting bacterial BGCs [5], [6] and shown to perform well. Supervised learning methods however are developed primarily based on annotated datasets, for which all instances are labeled as belonging to a specific class.…”