TechSprouts Monthly by Ankur Capital: January 2023

Ankur Capital
6 min readFeb 1, 2023


A monthly round-up on the deep science tech ecosystem in India

A peek into the future-Generative AI in biology, organic crystals and much more!

Image courtesy DALL.E 2

Deep science funding updates

  • Advanced battery solutions company Log9 Materials raised over $40 million as part of its Series B to expand manufacturing and R&D efforts. The round was led by existing investors in the company, Amara Raja and Petronas.
  • Chara Tech, building rare earth-free electric motors, raised $4.75 million in a round led by Exfinity Venture Partners. Their motors will be used in electric vehicles with further applications in HVAC systems.
  • Synthetic biology startup D-NOME announced its seed fund of $1.5 million, with investment coming in from Ankur Capital, Campus Fund and other investors. D-NOME is developing rapid diagnostic tools for point-of-care diagnostics with further applications in genomics.
  • Helex Bio, building a platform that makes genome editing safer, has raised $1.1 million as seed funding from IndieBio. The company is headquartered in New York but also operates a lab in Hyderabad.
  • Qritive uses AI to provide fast and accurate interpretations of pathology scans, primarily targeted towards cancer diagnostics and care. Headquartered in Singapore with a presence in India, it raised $7.5 million in a funding round led by MassMutual Ventures.

Deep science ecosystem updates

  • IIT Madras’ Gopalakrishnan-Deshpande Centre for Innovation and Entrepreneurship (GDC) hosted a seminar to generate ideas and paths forward to commercialize translational research and support deep-tech startups. So far, over 300 deep-tech startups have gone through GDC’s intensive boot-camps.
  • The Department of Science and Technology (DST), in partnership with IIIT Hyderabad and Microsoft, has announced a geospatial hackathon to build out solutions leveraging maps and satellite data to solve India-specific problems. The hackathon has two tracks: a startup challenge as well as a research challenge.
  • The India International Science Festival took place over four days in Bhopal. It featured 14 thematic events, which included school students, scientists, deep-science startups, and other ecosystem participants.
  • The Startup India Innovation Week took place in January 2023, with a number of entrepreneurship-focused events being held at cities across India. This included some deep science focused startup pitch events held at C-CAMP, Bangalore Bioinnovation Center.

News from the research community

  • Researchers at IISER Bhopal and the University of Queensland have developed a new flexible organic crystal (4-trifluoromethyl phenyl isothiocyanate) which shows promise in being used for high-sensitivity pressure sensors. Organic crystals have great potential in the sensor space as they are flexible and have low levels of defects compared to conventional electronics.
  • A team of researchers at Jadavpur University has discovered a technique to detect the level of pollution in an area by collecting and testing roadside dust for its magnetization. The technique enables the creation of a coarse but indicative map of pollution in an urban or industrial area.
  • IIT Tirupati and Kyndryl, an IT services company, have announced a collaboration to quickly develop cost-effective 3D prototypes for complex products in manufacturing. The collaboration will leverage IIT Tirupati’s domain knowledge in 3D printing and Kyndryl’s expertise in AI.
  • A team of scientists from Bengaluru’s Jawaharlal Nehru Centre for Advanced Scientific Research (JNCASR) have developed an artificial brain-like computing platform, based on scandium Nitride (ScN) tech. The platform offers high speed and low power consumption, and can be easily integrated with currently existing CMOS hardware.

Deep Science Thoughts

Generative AI in biology

Chat-GPT is all the rage these days. It’s like my personal assistant, my personal J.A.R.V.I.S!. It is built on top of Generative Pre-Training Transformer 3, or GPT-3, which is a generative AI language model heralding a step jump in the power of natural language processing. What sets it apart is its ability to converse with the user by understanding the context of the conversation, weigh the importance of different words and phrases and generate a response. While generative AI in language processing has seen huge progress, its application in the field of biology has also seen a steady growth.

One of the key challenges in the use of computational tools in biology is the inherent complexity of the systems in question. Take for example, proteins: they have a specific 3D structure/folding which gives them their functionality. The folding of the protein provides layers of structure built on top of its underlying amino acid sequence, which in turn is dependent on the DNA sequence which codes for the amino acids. There are two major problems in the modern study of proteins: predicting the structure of proteins from the underlying amino acid sequence, and secondly associating the structure to the right functionality. The difficulty is compounded by the size of the protein molecules. To solve these problems, large data sets are required which can be used to train specialised algorithms and generate predictive outcomes which are validated through wet lab experiments. The experimental data is fed back into the algorithms to improve their performance. This approach, coupled with sophisticated computational techniques, has shown promise and proved to be highly successful in the recent past.

AlphaFold from DeepMind and RoseTTAFold from Institute for Protein Design (IPD) at the University of Washington were among the first few algorithms to solve the first part of the problem: deducing protein structure from an amino acid sequence. AlphaFold2 currently has predicted 200 million protein structures with very high accuracy. In November 2022, Meta AI released the structures of more than 600 million proteins in the ESM Metagenomics Atlas database. Meta’s approach uses generative language models to generate its structures, which is essentially the same AI framework used by GPT-3. AlphaFold, on the other hand, uses pattern recognition tools such as transformers to identify interactions between amino acids at long distances in protein structures. While Meta’s algorithm isn’t as accurate as AlphaFold, it is significantly faster. Together, these structures cover the entire “protein universe” and provide the structure of almost every protein known to science.

The second part of the problem, namely associating a protein’s functionality to its structure, is significantly more difficult. However, generative AI tools promise to revolutionize this space too. Built on training through large data sets on protein structure-function relations, generative AI has the ability to design a novel protein with the desired application from scratch. Just this month, Nvidia along with Evozyne announced the development of two new proteins for specific functionalities, namely to reduce CO2 and to cure congenital diseases. This was done using a generative AI model based on the BioNeMo language training model. It followed the announcement by Absci who developed a zero-shot generative AI model to design novel antibodies in E. Coli, which means the model designed new protein molecules which were not been observed during its training. A few other companies in this field include Cambrium, Biomatter Designs, Cradle, Arzeda and Basecamp Research. AlphaFold itself is currently able to determine the functionality of single domain proteins but larger protein complexes are still a long way from being addressed. Another approach being considered is a combination of generative AI and AlphaFold. Insilico Medicine used a combination of its Pharma.AI platform along with AlphaFold to generate a new drug candidate for primary liver cancer. Solving this problem is still in its beginning stage and there is a long way to go before AI techniques can reliably generate proteins with the desired functional specificity. However, current methods along with advances in AI models show great promise in solving this puzzle.

The ability to design proteins with specific functionality can open up huge opportunities in drug discovery as well as diagnostics. Use of generative AI can reduce the timelines for new drug discovery and reduce the cost of manufacturing as well. It can also open the field of precision medicine in the future. As the technology matures and specific applications in biology are explored further, generative AI applications in biology are expected to attract significant venture capital investment. The science of proteins marks one of the frontiers of modern biology; the use of computational tools not only promises a variety of applications, but also gives us a clearer understanding of the fundamental nature of proteins.

To receive this newsletter via email, subscribe here: TechSprouts.