Team:IISER-Pune-India/Contribution
New documentation for existing parts
We worked with the biobricks BBa_K1362400 (N-Intein), BBa_K1362401 (C-Intein), BBa_K1362424 (Bsa1 recognition site) for our project. These biobricks were designed by Team Heidelberg 2014. They were part of the system they had designed as a circularisation construct (BBa_K1362000) that could be used to circularize any protein. You can find the link to their construct here. We verified the circularisation construct with the BBF RFC [105] reference file to ensure that all the required sub parts were present. We picked these biobricks as our team is working to develop a peptide drug consisting of an inhibitory drug engineered into the scaffold of a circular protein called a cyclotide. For this purpose, the desired inhibitory sequence was grafted into loop 6 of the cyclotide. We are specifically working with the cyclotide Kalata B1. While trying to incorporate these into our project, we came across certain problems. On examining these biobricks, we found that these biobricks were not suitable for restriction free cloning/ Gibson’s assembly. We mentioned these problems on the main pages of the respective pages on the registry and modified them by making them in frame with other components thereby making it compatible for Gibson’s assembly/RF cloning. You can find our modified biobricks here - BBa_K3582020, BBa_K3582022 and BBa_K3582021. These improved biobricks can be used by future teams who can incorporate these parts into their own projects, especially while designing their own parts. The modified inteins are compatible with restriction-free cloning/ Gibson’s assembly technique and are readily codon optimised with E-coli strain K12 . The modifications we made were the following:
Replaced the RFP protein with the Cyclotide of our interest.
Removed the T7 RBS, the Scar and the RFC[105] start codon from the start of the linear construct. The gene sequence then begins with the sequence that codes for NpuDnaE (C) Intein.
At the end of the construct, we removed everything after the two stop codons [TAATAA] - Keeping the stop codons was a necessary condition in accordance to the RFC[105] standard.
Removed the 7th nucleotide from the two Bsa1 recognition sites that flank the insertion site. The removal of these two nucleotides fixed the ORF.
Combining an already existing software and hardware tool and building up on the software
To come up with a complete end-to-end diagnostics solution for malaria detection we are using a low-cost paper-centrifuge, a foldscope that is used to capture blood smear images of patient blood samples, and a Deep learning application that classifies blood smear images and predicts the probability of the image being from an infected patient.
Hardware
The foldscope was previously used by Team Lambert_GA 2015 , where they designed filters to turn the foldscope into fluorescent microscopes. The foldscope was also used by Team TUDelft 2019. They held a foldscope event in which they instructed the participants on how to fold their own origami microscope and how to use them. We have contributed to and refined these projects by pairing the tool with a machine learning software and creating a user manual for the diagnostic kit. This would allow virtually anyone to use the kit, without the consultation of healthcare professionals and experts. The manual contains video tutorials and demonstrations for a better understanding of how to use the kit.
The Foldscope was invented by Prof. Manu Prakash and Jim Cybulski at Stanford University. The Paper Centrifuge has been developed as an ultralow-cost way of solving the critical bottleneck caused by expensive, bulky and electricity powered commercial centrifuges. The software for analysing the blood smears was developed by IISER Pune’s iGEM team based on deep learning models for image recognition. The Foldscope and Paper centrifuge replaces the need for a laboratory while the deep-learning software replaces the need for an expert to identify infected RBCs.
Software
We wrote a web scraper than parses through the RCSB Database and returns a single .csv file with all the information requested. This parser can be further optimised by using multi-threading and multi-processing by other teams to improve the rate of I/O transfer and preprocessing to scrap large amounts of data.
Team Heidelberg 2017 developed DeeProtein, a deep learning framework based on ResNet50 architecture for protein function prediction from raw sequence data. We used the same residual neural network architecture (particularly the 30 residual blocks) and optimized it for classifying malaria blood smear images. Furthermore, we modified the top layer and added a single dense neuron for Binary classification while DeeProtein was used for Multi-class classification. We trained this Deep neural network for binary classification on a GPU for many days and achieved an accuracy of ~95% on our Dataset. Essentially, we used the Base model of the Network they built, modified it for our dataset and implemented it.
This network along with its weights can be used by other iGEM teams in the future for computer vision applications. We have provided a tutorial and a handbook for future teams on our wiki and Github repository. Our model serves as a starting point for the interdisciplinary nature of methods required for tackling problems in medical diagnosis and imaging. In the next year of iGEM, we hope to improve upon our current model and soon build a comprehensive mobile device that is trained on more pictures taken from a foldscope.