_uploads__1596694138_big_presentation_format.pptx

This is a presentation on detection of AMR in tuberculosis and a business model around it

1. Development of NGS driven Machine Learning based rapid diagnostic platform for Antimicrobial resistant tuberculosis Submitted by Dr. Animikh Ray

2. Problem/Need Multi-drug resistance is a major cause for concern in LMICs like India and tuberculosis is a disease that has been a focus area for Government of India for eradication Tuberculosis affects healthcare system of India adversely with disproportionate number of patients belonging to below poverty line The rise in antimicrobial resistance (AMR) is endangering effectiveness of antibiotic treatments, leading to possible therapy failures. Given the growing threat of AMR and reduced antibiotic research and development (RD) there is a need for further research for tools such as interpretable machine learning to predict resistance and steer effective treatments for ailments like tuberculosis

3. Opportunity culture and isolation of microorganisms are considered a primary traditional diagnostic method. However, this process is time-consuming, as it takes a considerable amount of time to obtain culture results. Improper cultivation of microbes also increases the risk of obtaining false negative results. Moreover, the excessive use of antibiotics in clinical practice can further hinder the growth and isolation of specific organisms . This proposal aims to utilize the very nascent use of large language models in genome sequence data analysis in order to develop an innovative point of care diagnostic tool that will allow effective clinical decision support for determining course of therapy for patients infected with drug resistant Mycobacterium tuberculosis The diagnostic tool will ascertain the drug resistance pattern and provide rapid diagnosis for patients in order for the clinician to make an informed decision regarding course of therapy.

4. Define PoC <18 months> The diagnostic tool that has been developed at Father Muller Medical College and research center in collaboration with National Institute of Technology Karnataka at Surathkal , and CDAC,Govt of India will be field tested using clinical isolates obtained from patients at multiple sites including of Father Muller Medical College and research center, Wenlock Government Hospital district TB center and Kasturba Medical College and Hospital, Manipal so that TRL-6 can be achieved.

5. The proposed solution Technical Details (1/2 ) The model is designed based on nucleotide LLMs Nucleotide LLMs, using DNA/RNA alphabets. The initial model has been designed based on CARD database which is a comprehensive database of resistance genes, their products and associated phenotypes. The LLM that was designed leveraging genome-scale data to model individual mutations at the nucleotide scale, thus implicitly accounting for protein-level mutations at the codon level. The foundation model was designed on a wide set of parameter scales ranging from 25 million to 25 billion, with a maximum sequence length of 2,048 tokens. The descriptor set used for the generation of the models included sequence and structure based features available in the CARD database. A subsequent use of the models will be its ability to generate new drug resistant sequences, with the eventual goal of predicting yet unseen drug resistance pattern. This model was tested on 200 Indian Clinical isolates despite inherent heterogeneity in strains. In This proposal the model will be clinically evaluated in multiple sites to achieve TRL-6 and compared with Traditional antibiotic susceptibility testing using UKMYC6 broth dilution plate and Lowenstein-Jensen proportion method.

6. Technical Details (2/2) (Or Supporting Evidence) For DNA sequences, a one-hot encoding scheme is used, where each nucleotide base is represented by a binary vector: Adenine (A) → [1, 0, 0, 0] Cytosine (C) → [0, 1, 0, 0] Guanine (G) → [0, 0, 1, 0] Thymine (T) → [0, 0, 0, 1]

7. MODEL ARCHITECTURE Sequential Model Dense Layer Activation Function: Relu ( For Non Linearity) Loss Function

8. The high testing accuracy reaffirms the reliability of the model in real-world applications. The model's performance provides confidence in its utility for predicting resistance mechanisms and medicine classes accurately. This milestone sets a promising foundation for further advancements in combating antibiotic-resistant bacteria.

9. Dr. Animikh Ray Dr. Anwesha Chatterjee Dr. Mohammad Rizwan Dr. Shyam Lal Mr. Abhishek Mr Vishal Mr Anindya Kundu This is a multidisciplinary team consisting of basic biological and computer science researchers as well as skill programmers as well as corporate data scientists and business analysts responsible for successful execution and commercialization of the product

10. Novelty No such whole genome sequence data driven large language model based diagnostic tool exists for bacterial antimicrobial resistance detection in general and tuberculosis in particular. Large language models are a nascent paradigm in machine learning and this paradigm has not been applied for whole genome sequence data analysis. Diagnostic tool proposed in this work does not exist in India or globally. Several companies have similar product for oncology ( qiagen , mapmygenome ) but non for infectious diseases, AMR or tuberculosis.

11. Feasibility With increasing efficiency of reduction in cost of genome sequencing this proposal is entirely feasible as it has already achieved TRL-3 and we have tested on whole genome sequence of clinical isolates in North India. IP Strategy Once the diagnostic platform achieves TRL-6 intellectual property protection process will commence.

12. Market size

13. Competitive Landscape   Bacterial genome sequence Customised bioinformatics pipeline LLM based clinical decision support tool Time taken for report generation Our Diagnostic tool Yes yes yes 1 working day CosmosID® (USA) Yes no No 3-5 working days illumina Yes Yes No 3-5 working days Qiagen Yes No No 3-5 working days Oxford Nanopore Yes Yes No 3-5 working days

14. Define Differentiation of Proposed Solution No other diagnostic platform uses LLM based deep learning paradigm in reporting antimicrobial susceptibility in tuberculosis No other platform caters to clinical healthcare providers for design of therapy No such clinical support decision tool exists Mention what is the Value Proposition Our solution is cheap with faster turn around time that may be crucial for critical care patients and reduce therapy burden. Tuberculosis patients have to take tremendous amount of medication leading to patient non-compliance and mortality. This tool will prevent that.

15. When do you plan to set up your startup ? Following successful completion of BIRAC BIG grant an entity will be created for commercialization

16. Have you discussed about your solution with Target Customers? How many >100,>50,>25,>10,>5, None? What is the feedback? NA Have you discussed your solution with business mentor? What is the feedback? This can be a successful Enterprise solution that can integrate multiple facets through the interchange of information from various process areas and related databases . If executed carefully this can potentially be an unicorn in the diagnostic domain.

17. Projected Time to Hit the Market 3- 5 yr Road Map Tasks (months) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 24 30 36 48 60 purchase of chemicals Recruitment of human resource Development and evaluation of POC(TRL-4) evaluation of POC with larger sample size at one center (TRL-5) evaluation of POC with larger sample size at multiple centers (TRL-6) Patent and intellectual property protection Fund raising Development and integration with channel partners Expansion into market outside India

18. Business Plan • Target Customer • Revenue Model Target Customer: B2B: Hospitals , diagnostic chains B2G: state and central governments

19. Work plan (Gantt chart) Tasks (months) 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 purchase of chemicals Recruitment of human resource Development and evaluation of POC(TRL-4) evaluation of POC with larger sample size at one center (TRL-5) evaluation of POC with larger sample size at multiple centers (TRL-6) Patent and intellectual property protection

20. Milestones Duration (months) Milestone targets 3 Recruitment and protocol optimization 6 POC validated in laboratory scale(TRL-4) 9 evaluation of POC with larger sample size at one center(TRL-5) 12 TRL-5 completed 15 evaluation of POC with larger sample size at multiple centers(TRL-6) 18 TRL-6 completed

21. Budget Heads Amount (Rs lakh) Basis Equipment Consumables 12 For laboratory work Man power 15 lakhs For design of Machine learning model and bioinformatics pipeline Incubation services Travel 3 For field work External services 15 For sequencing related services Contingency 5 For execution of alternate sequencing protocol and machine learning infrastructure