Principal Investigator

About the role

# Principal Investigator, Biological Computing frontier · London / Cambridge, UK · Full-time, permanent Founding scientific leadership role (CTO/CSO equivalent) Salary: £100,000 to £180,000 (approx. US$136,000 to US$246,000), plus 1 to 4% equity --- ## About frontier computing frontier computing works on cultured neural tissue as a computational substrate. The underlying observation is not new. Biological neurones in vivo perform inference at very low power, several orders of magnitude below what a comparable silicon system would draw, and the cells manufacture themselves. What has been missing is any serious attempt to find out whether that can be turned into something reproducible, scalable and cheap enough to be useful. For most of the past decade the answer was clearly no, for well understood technical reasons. Several of those reasons have weakened: iPSC differentiation protocols have become more reliable, microfluidic culture systems now survive well past the four week mark, and reading from and writing into cultured tissue with multi-electrode arrays is better understood than it was. We should be plain about where we are. We have learning on a very basic substrate, and a strongly supported hypothesis that further scaling of substrate size will bring further computational results, enabling energy efficient computation at scale while circumventing von Neumann architecture bottlenecks. That is a starting point, but there is significant the work ahead is to establish whether the mechanisms we are seeing and reproducible, generalise to more structured circuits, and survive being scaled. Credit assignment is an incredible challenge and an open question within biological neural networks, and one that our company is actively solving. --- ## The research programme Three codependent problems define the work. ### 1. The cost of culturing substrates at scale Large scale neuronal culture is currently expensive enough that very few groups can do sustained work with it. We think that is the binding constraint on the whole field, and that it is more tractable than it looks. The useful denominator is cost per neuron-hour rather than cost per millilitre of media, which decomposes multiplicatively into chemistry, volume and density. Within chemistry the drivers are not where one might expect: in a representative maintenance medium the basal medium and the B-27 supplement together account for the large majority of per-millilitre cost, while the recombinant factors, which are by far the most expensive per milligram, contribute very little because they are dosed at trivial mass. Both of the dominant inputs are commodities, and commodities respond to process work rather than to chemistry. Volume and density compound on top of chemistry, and have received less attention, largely because the protocols in general use were developed for developmental biology rather than for sustained culture. There is meaningful headroom across the product of the three. We have made real progress here, and it is the most mature of the three programmes. The goal is not a proprietary advantage in reagent purchasing. It is to bring large scale neuronal culture within reach of ordinary experimental budgets, including our own. ### 2. Reproducing core neuronal training mechanisms Most work in this area imports the reinforcement learning apparatus from silicon: define a global reward, propagate it back, and hope the tissue converges. This is expensive to build and expensive to run, and it scales poorly as neuron count rises and credit assignment gets harder. We are interested in the opposite approach, which is to let the biology do as much of the learning as it already knows how to do. Local mechanisms, spike-timing dependent plasticity and neuromodulator-gated plasticity among them, are implemented natively by the tissue at no cost to us. If a training environment can be designed so that the desired change is what those mechanisms produce anyway, the control apparatus becomes much simpler and the approach scales with neuron count rather than with the complexity of the reward circuitry. Our current work is in **cortical computation**, and the near-term direction is to extend it toward **hippocampal circuit work**, where the relationship between architecture, plasticity and function is better characterised and where there is a substantial literature to argue with. The immediate scientific question is whether the learning we see on a simple substrate reproduces in a structured one. ### 3. Building out mature neuronal cultures at scale Culture size is limited by oxygen and nutrient diffusion, and beyond a few millimetres the core of a construct necroses. This is a well known constraint and it is the proximate reason nobody has built a biological system with the connected mass that would make any of this interesting. Maturation is the second half of the problem and is often treated separately, which we think is a mistake. A large culture of immature neurons is not useful; the target is tissue that is both sizeable and functionally mature, held in that state for the duration of a workload rather than for the duration of an experiment. Vascularisation is the approach most of the literature takes, and progress there is real. We are interested in whether it is the only approach, and our roadmap is not contingent on it. We would be glad to be argued with on this point by someone who knows the perfusion literature better than we do. --- ## About the role This is the founding scientific leadership position at frontier. We have called it Principal Investigator in this job posting rather than Chief Scientific Officer because the substance of the job is closer to running a research group than to running a corporate function: you would set the scientific agenda, choose the problems, build the team, and be accountable for whether the science works. The title officially will be CSO, but your functional work will have much more freedom than this role traditionally implies, given the greenfield nature of the field. However, your work will fundamentally be application-oriented, designed to push the capacity for biological substrate approaches. This is not a role performing blue-skies research, and the programme will have to be justified towards our business goals. First and foremost, we are looking for someone who understands these tradeoffs, and appreciates the science they are pursuing must benefit our commercialisation strategy: building larger tissues with both greater and denser computational and memory capacities. You would be the first senior scientific hire, and as part of that role you will define the future composition of the team alongside the CEO, and help buildout a scientific programme that drives substrate progress and feeds into our commercialisation efforts. ### What you would be doing Research leadership. Own the scientific agenda across the three programmes above. Decide what is attacked, in what order, with what resources, and decide what is stopped. Design the experimental programme that takes us from the substrate we have now to something considerably larger, with the network kept in the loop rather than distilled away. Building the group. Recruit and lead the founding scientific team across wet lab, instrumentation and the computational side of the training work. You would have real latitude over who joins and how the group is organised. If you hold views about how research groups ought to be run, this is an unusual opportunity to act on them. Technical judgement. Decisions about tissue configuration, media formulation, electrode interface and training environment interact, and have to be made as a set rather than separately. This is the part of the job that cannot be delegated and is the reason the role exists. Scientific representation. Publish, present, and represent the work to the field, to collaborators, and to the people who need to understand why this is a long horizon programme rather than a short one. Much of the work that you are leading will be obfuscated within the short term, while we scale the platform and create a technical moat around our outputs, but it's vital that we share the fruits of our work with the scientific community as much as can be reasonably permitted. --- ## What we are looking for A PhD in neuroscience, bioengineering, tissue engineering, biophysics, electrical engineering or a related field, together with a record of independent work: postdoctoral, fellowship-funded, group leader or industrial. We are interested in what you have built and what you understand rather than in how long you have been doing it. Substantial hands-on expertise in at least two of the following, and the ability to reason about the whole system: * Neural cell culture. iPSC differentiation to cortical or other neuronal identities, long duration culture, media formulation, and the practical business of keeping tissue alive and functional over months. * In-vitro electrophysiology. High density and CMOS microelectrode arrays, closed-loop stimulation and recording, spike sorting and population analysis at high channel counts. * Tissue engineering and microfluidics. Three dimensional culture, perfusion, scaffold and substrate design, and a working understanding of designing culture substrates to circumvent the mass transport limits that govern construct size. * Computational modelling of neural systems. Spiking network models, plasticity rules, and the analysis of learning in biological networks. We would particularly like to hear from you if you have: * Worked on plasticity or learning in cultured or ex-vivo tissue, especially with local rules rather than externally imposed reward. Experience within biocomputing on neural organoids and eliciting learning in these tissues is particularly beneficial. * A background in hippocampal circuit physiology, or in the relationship between neuronal circuit architecture and memory more generally. * Taken a biological protocol from research grade to something reproducible and cost controlled, through process development, QC, automation or in-house reagent preparation. * Worked on maturation or on the size limit in three dimensional neural constructs, and formed a view about it. * Built instrumentation across electrophysiology, optogenetics, or other related areas, as well as used it. A note on temperament, since it matters here as much as technique. This work suits people who are comfortable being among the few in the world attacking a particular problem, who will stop their own project when the data says to, and who are willing to treat a cost model as a scientific object rather than an administrative chore. It suits people who are not troubled by a thin literature. It is a poor fit for someone who needs an established field to orient within, and we would rather establish that early than late. We are not looking for someone whose work primarily applies existing organoid culture or MEA tools as instruments. The contribution includes the design of the system itself, recognising the importance of co-development of the technology substrate with the --- ## What we can offer Compensation. £100,000 to £180,000 (approximately US$136,000 to US$246,000 at current rates) depending on experience, plus **1 to 4% equity** on a standard four year vest with a one year cliff. The range is genuine and negotiable, and it reflects that this is a founding position. You would be an owner rather than a member of staff. Scientific authority. You choose the problems and you stop the projects. The founder's job is to make sure the company survives long enough for your programme to run. Yours is to make sure it is the right programme. Resources. Recurring biology, wet lab and microfluidics infrastructure are funded, and you would shape what we buy rather than inherit someone else's equipment. Budget for the founding scientific team is yours

Similar positions