Last January, German biotech company BioNTech acquired African AI startup Instadeep for more than $550 million, with the deal completed in July of the same year. InstaDeep's exit from Africa is currently the largest, and it has only been operating for just over a year under the umbrella of a German pharmaceutical company. Now is a good time to look back at the company's performance since the acquisition.
Instadeep uses advanced machine learning techniques to bring AI to enterprise applications. Its products range from GPU-accelerated insights to self-learning decision-making systems. Prior to last year's acquisition, the Tunis-born, Paris, London-headquartered enterprise AI startup had raised more than $108 million from multiple global investors, including Google, Deutsche Bahn, and BioNTech. These three strategies were also the startup's biggest partners and customers.
Notably, the decade-old startup has teamed up with BioNTech to develop an early warning system that can detect high-risk COVID-19 variants months in advance during the pandemic. Instadeep collaborated with Google DeepMind to create an early detection system for desert locust outbreaks in Africa. We also collaborated on the Moonshot project to automate train schedules for Deutsche Bahn, Europe's largest railway operator.
While these partnerships demonstrate a variety of applications for Instadeep's solutions, the company's acquirers had clear use cases. The company uses AI to develop treatments and vaccines for various cancers and infectious diseases, and is now ramping up its efforts under new ownership.
Fifteen months after completing the BioNTech acquisition, co-founder and CEO Karim Begil said in an interview with TechCrunch that while Instadeep is an independently operating AI company, it still provides solutions to external clients. Despite this, he said he has made great progress in that area. biotechnology.
“We are strategically aligned with BioNTech regarding the goals we pursue with our biology and bioAI capabilities,” the Instadeep chief said. “But we also have scope to continue to be a force in Africa and in the AI space in general, while continuing to develop technology that pushes the frontiers of innovation in other areas such as industrial optimization.”
Improving capabilities in biotechnology
Beguir notes that the goal in the year since the Instadeep acquisition has been to bring AI to every stage of BioNTech's pipeline to improve existing processes.
He provides an example from histology. This involves tissue analysis and visual work to label different tissues, such as identifying tumor cells or healthy cells. According to him, BioNTech experts traditionally did this work manually. However, Instadeep's technology accelerated the process by introducing visual AI and segmentation systems, making this labeling organization workflow 5x faster.
The other is the completion of the RiboMab project. The project involves mRNA-encoded antibodies that are now part of BioNTech's toolkit as an immunotherapy company to fight cancer and other diseases. In its first collaboration in 2020, InstaDeep introduced this project on its DeepChain platform for designing proteins and analyzing biological data.
Biotechnology involves a wealth of sensitive medical data. Collecting and analyzing them is another thing. Keeping them safe is another thing. Just ask 23andMe. Once hailed as a disruptor in the biotech space, it has since fallen victim to a massive breach that compromised the data of nearly 7 million people, half of its customer base.
Interestingly, BioNTech is no stranger to such events. In 2020, hackers attacked the European Medicines Agency (EMA), Europe's medicines regulator that evaluates medicines and vaccines, and illegally accessed documents related to the company's coronavirus vaccine developed with Pfizer. While Pfizer and BioNTech have confirmed that their systems and trial data remain secure, the incident highlights how vulnerable organizations are to cyberattacks, even regulators. .
As any CEO would tell you, Beguir told me that Instadeep and BioNTech are very cautious about healthcare data. In particular, the partnership is now using AI to increase its data assets, which could identify precise protein sequences and uncover new targets for cancer and other immunotherapies. Usage example.
However, there is granularity in the data used by both companies. BioNTech works with real patient personal data, and Instadeep typically develops and trains models on publicly available data. This is, for example, how we trained the Nucleotide Transformer, a suite of AI genomics models that is now the most downloaded and popular AI genomics model in the world. [Thanks in part to this open-source deal.]
“Instadeep developed and trained a nucleotide model based on publicly available data,” Beguir said. “However, when we want to deploy our models to specific use cases or real patient data, we can benefit from all the privacy guarantees that come from our position as one of the leading biopharmaceutical companies operating under strict controls. We have done this at the BioNTech level following regulations and strict quality protocols.
Development of new technologies within BioNTech and outside BioNTech
When asked what Instadeep's next milestone will be within BioNTech, Beguir mentioned the startup's “latest breakthrough”: the Bayesian Flow Network (BFN). This is a new protein generation AI model that significantly outperforms autoregressive and diffusion models. BioNTech CEO Ugur Sahin described BioNTech as “cutting edge technology” in a statement.
According to Beguir, the model allows the system to search for specific properties on an antibody's heavy chain, such as chemical properties, hydrophobicity, and sequence length, to find the most natural-looking, best-behaving proteins on the market. It is said that it will produce. Such models are critical for understanding complex protein function and designing new therapeutic proteins.
“We are committed to using AI like ours to identify real-world use cases, work closely with BioNTech, and build products that will be tested in labs and clinics, and ultimately save patient lives. We are excited about the possibilities for innovation,” said Beguir. “If you think about the state of biology and AI today, it's similar to the state of natural language processing with GPT-3 in 2020. The system is starting to work, and it's been great, but there's still room for improvement. There was.”
Instadeep announced a new AI model last week, along with a new near-exascale supercomputer, and the companies say the partnership will put it in the top 100 of global computing and infrastructure clusters, as well as the top 20 H100 GPU clusters.
Both developments focus on where BioNTech-owned Instadeep is deploying AI in several life sciences use cases. Meanwhile, we also independently conduct other businesses related to AI and deep reinforcement learning for industrial optimization.
One example is a 12-year ongoing project to automate train planning and dispatch for Deutsche Bahn, one of our long-standing partners and Europe's largest rail operator. Similarly, the Tunis- and London-based AI company is working to develop other industrial optimization use cases, such as collaborating with Germany's Fraport to use AI to optimize complex airport operations. We are strengthening it.
“In general, we see the promise of AI agents as very attractive. We believe that industrial optimization and agent-based systems that work in conjunction with human colleagues will help industrial We believe it will revolutionize efficiency. So this is another area that we have been working on for many years and one that we continue to invest in,” Beguiar said.
Meanwhile, earlier this month, Instadeep launched the professional version of its DeepPCB product, a hardware or printed circuit board design fully backed by autonomous AI powered by reinforcement learning, in San Francisco. Beguir said the company's competitors are smaller AI startups in the niches in which it operates, such as Riyadh-based Intelmatix.
Instadeep leaders say their company is moving away from simple things like Gen AI for NLP and solving more complex AI use cases, such as Gen AI for DNA and proteomics and agent workflows for combinatorial optimization. We are proud of our products. He argues that this ingenuity, in addition to the BioNTech acquisition, has played a considerable role in driving inbound interest from customers in the US, where the AI company currently has two offices, and also across Europe. , which also claims to have locations in Berlin, Paris and the UK, among others.
BioNTech spent $500 million on Instadeep to enhance its biotech capabilities, but for this reason the AI company's operations will continue to grow while funding activities that serve customers beyond the biotech industry. maintains its independence. ”
When asked why BioNTech still allows AI companies to work on non-biotech projects, it says, “We are contributing value by being a leader in AI, and AI skills can improve across multiple sectors. ,” Beguir replied. “With the same technology stack, no time is wasted working on non-biotech AI. BioNTech also deploys InstaDeep for tasks outside of biotech R&D, such as operational optimization.”
Beguir said that while InstaDeep was not forced to sell, the company's shared vision and successful projects with BioNTech since 2019, long before the acquisition, convinced the AI company to move forward with the deal. Explain that it is. He believes that the trust built through years of collaboration is the reason InstaDeep remains independent under BioNTech. Now, it's important for InstaDeep to maintain its momentum, maintain its high-quality results, and continue to innovate for as long as possible. ”
Since the acquisition, InstaDeep has grown to over 400 employees worldwide. This includes an African team based in a new office in Kigali that will lead the company's geospatial intelligence operations.
Originally a ground-based effort in partnership with Google to detect locust breeding grounds in Africa, Instadeep is now using historical label data and satellite imagery to discover where locust breeding grounds are located in Africa. We estimate it with high quality and 80-85% accuracy. next 30 days. Beguir said the company's framework, InstaGeo, which uses multispectral satellite imagery from NASA and the European Space Agency (ESA), is open source and can be used by other companies to develop solutions that are scalable across the continent. He says it can be done.
“This is a real-world example of how AI technology and capabilities are having an impact. Rather than collecting samples on the ground or relying on terrestrial infrastructure, these insights can be delivered via satellite. at scale and inform multiple governments and stakeholders to address the growing challenges to food security, especially given the continent's climate challenges.”