The use and superior performance of machine learning to several complex tasks like image classification and object recognition inspired us to explore the applications of machine learning in combating the problem of antibiotic resistance. We identified several applications like the prediction of the phenotype of strains, protein-protein interactions, and finally decided to explore and understand the genetic mechanism behind antibiotic resistance.
We have analyzed the results of our machine learning extensively and concluded that our approach has the potential to impact Research Community and also, our own iGEM project.
Our machine learning algorithm is able to detect genes that confirm existing data and experimental evidence, which is already validated by the experiments. The detection of such genes validates the effectiveness of our approach. It is well known that aadB and rmtB are responsible for resistance to aminoglycoside antibiotics, which is also detected by our machine learning algorithm.
There are several genes that are detected by the machine learning approach which are not proposed or studied in the literature. These genes might have been mainly overlooked for the case of a particular antibiotic or even the whole organism. There are several genes like esiB, puuP, emrE, relE, tufA, yafQ, etc. which are detected by our machine learning algorithm and shall be tested for their importance in the case of A. baumannii.
These genes which are the most important features for predicting phenotype as analyzed by machine learning algorithm can be knocked out of the strain to make the strain susceptible to the particular antibiotic. We have also detected genes like msr(E) and bla, which play an important role in resistance in hospital settings and hence shall be knocked out of those strains.
Applicability to other pathogens:
The machine learning approach can be easily applied to any pathogen of interest. The major requirement is the presence of data having different strains and their resistant phenotype. We have provided the codes of the program, for other researchers to replicate the approach to other pathogens. The codes are available here
Extensive Computational Analysis:
The machine learning algorithm helps in knocking off important genes to make strain susceptible to antibiotics. In case, the strain is still resistant to antibiotics, we can treat that bacterial strain with our novel protein-based therapeutics. We have also developed another software, called TailScout for automating the process of predicting the secondary structure of our fusion protein for any gram-negative pathogen.
It would be interesting to test our novel engineered fusion protein for its efficacy against the resistant genes detected by our machine learning algorithm for various antibiotics. The performance of our novel protein-based drug against these genes would help us in validating the efficacy of our drug. The genes like aadB and rmtB are responsible for resistance to aminoglycoside, gene tufA has been shown to play the role of a virulence factor of A. baumannii, which makes it very important to test these genes with our novel protein-based drug.
As our approach has detected several novel targets or pathways, which are not yet proposed in the literature, it would be exciting to test the performance of our novel drug against these novel genes. We discovered several novel genes like eptA, relE, and yafQ which are not yet explored much in the case of A. baumannii therefore making it much more important to test with our protein-based drug.
ML x SynBio
Utilizing machine learning in our project this year has made us realize that machine learning has a very important role in the advancement of synthetic biology. The best example of this intersection is the knocking of resistant genes detected using machine learning from different strains of bacteria to make it more susceptible to antibiotics. Therefore, machine learning has guided us towards genetic engineering which is the basic step of synthetic biology.