Periodic Reporting for period 4 - Bio4Comp (Parallel network-based biocomputation: technological baseline, scale-up and innovation ecosystem)
Reporting period: 2021-01-01 to 2022-06-30
In Bio4Comp we improved an alternative, parallel computing paradigm in which a given combinatorial problem is encoded into a graphical, modular network that is embedded in a nanofabricated planar device. The mathematical problem is then solved by exploring the network in a parallel fashion using a large number of independent, biological agents, namely molecular-motor-propelled protein filaments. This novel approach – in the following termed network-based biocomputation (NBC) – has the critical advantages that it uses orders of magnitude less energy per operation than conventional computers and that it is scalable with presently available technology. Our vision is to use future NBC devices to solve specific problems, including nondeterministic-polynomial-time (NP) complete problems that are intractable with conventional, sequentially-operating electronic computers. Such a devices have the potential to greatly improve our ability to solve many problems of practical importance such as the de-coding of encoded messages, the design and verification of circuits, the folding and design of proteins and optimal network routing.
Main objectives
The overall aim for Bio4Comp was to establish the technological and mathematical baseline and the interdisciplinary innovation eco-system needed to systematically scale up NBC technology and thus to create the pre-conditions for applications. Specifically, we targeted the following breakthroughs:
1) A network-based biocomputer that solves a more complex problem than any other alternative parallel computing approach. This was enabled through targeted mathemati-cal, biological and nano-technological advances and their integration, to enable fully scalable, ef-ficient, parallel problem solving.
2) New networks that efficiently encode combinatorial problems of practical im-portance. To solve a specific type of problem, NBC requires an “algorithm” in the form of a network designed to encode the problem. The focus of our efforts was to efficiently and in a scalable manner solve the 3-SAT problem that is at the core of electronic computer design and mathematics.
3) A structured, interdisciplinary community with critical mass and an innovation eco-system for NBC development. Through a set of coordinated, targeted actions we aimed to create a structured community from basic research to end-users.
Main objectives
The overall aim for Bio4Comp was to establish the technological and mathematical baseline and the interdisciplinary innovation eco-system needed to systematically scale up NBC technology and thus to create the pre-conditions for applications. Specifically, we targeted the following breakthroughs:
1) A network-based biocomputer that solves a more complex problem than any other alternative parallel computing approach. This was enabled through targeted mathemati-cal, biological and nano-technological advances and their integration, to enable fully scalable, ef-ficient, parallel problem solving.
2) New networks that efficiently encode combinatorial problems of practical im-portance. To solve a specific type of problem, NBC requires an “algorithm” in the form of a network designed to encode the problem. The focus of our efforts was to efficiently and in a scalable manner solve the 3-SAT problem that is at the core of electronic computer design and mathematics.
3) A structured, interdisciplinary community with critical mass and an innovation eco-system for NBC development. Through a set of coordinated, targeted actions we aimed to create a structured community from basic research to end-users.
Regarding Aim (1), to solve a problem that is more complex than what any other alternative computing approach has solved, we defined the problem complexity as the number of potential solutions that need to be checked in order to solve the given instance by brute force. At the start of the project, the benchmarks were a 20-Variable 3-SAT problem with a complexity of ~1 million solved by DNA compu-ting in the year 2000 (red dotted line in Fig. 2) and the factorization of 143 with a complexity of ~250 solved by quantum computing in 2012 (yellow dotted line in Fig. 2). Our plan was to surpass the then state of the art in quantum computing by solving an Exact Cover instance with a complexity of 1000+ by mid-2019, and DNA computing by solving a 3-SAT instance with a complexity of 1500 000 by the end of 2021 (blue circles and dotted line in Fig. 2). We achieved these goals with some delays of about one year by solving 10-set Exact Cover instances by the end of 2020 and a 25-set Exact Cover in-stance by mid 2022 (blue circles and dotted line in Fig. 2). Note that the 25-set Exact Cover results are unpublished data presented in the Bio4Comp D4.6 deliverable report.
Regarding Aim (2), to efficiently encode combinatorial problems of practical importance, we significantly optimized the network encoding for Exact Cover and developed two 3-SAT network encodings: (i) the space-encoded SAT network algorithm that is conceptually similar to the existing Subset Sum and Exact Cover network algorithms, and (ii) the agent-encoded SAT network algorithm that enables very compact networks but requires the ability to store information on the agents themselves. This is a key step towards broad general applicability of NBC because it enables solving NP-complete problem for which a polynomial conversion to SAT ex-ists. Furthermore, this result provides a novel NBC approach that is significantly more scalable because the networks are much more compact. In particular the agent-encoded approach – once demonstrated experimentally – holds promise for future applications of NBC, because it allows compact networks that scale linearly with the number of variables and are independent of the number of clauses in the 3-SAT instance. Furthermore, such networks can be used to solve many different SAT instances.
Regarding Aim (3), to create a structured community for NBC development, our most valuable outcome is our Roadmap for network-based bio-computation, which will help to “identify pre-competitive research domains, enabling cooperation between industries, institutes and universities for sharing R&D efforts and reduction of development cost and time”. Our ability to perform the originally planned stakeholder workshops was substantially limited by the pandemic that started in early 2020.
Regarding Aim (2), to efficiently encode combinatorial problems of practical importance, we significantly optimized the network encoding for Exact Cover and developed two 3-SAT network encodings: (i) the space-encoded SAT network algorithm that is conceptually similar to the existing Subset Sum and Exact Cover network algorithms, and (ii) the agent-encoded SAT network algorithm that enables very compact networks but requires the ability to store information on the agents themselves. This is a key step towards broad general applicability of NBC because it enables solving NP-complete problem for which a polynomial conversion to SAT ex-ists. Furthermore, this result provides a novel NBC approach that is significantly more scalable because the networks are much more compact. In particular the agent-encoded approach – once demonstrated experimentally – holds promise for future applications of NBC, because it allows compact networks that scale linearly with the number of variables and are independent of the number of clauses in the 3-SAT instance. Furthermore, such networks can be used to solve many different SAT instances.
Regarding Aim (3), to create a structured community for NBC development, our most valuable outcome is our Roadmap for network-based bio-computation, which will help to “identify pre-competitive research domains, enabling cooperation between industries, institutes and universities for sharing R&D efforts and reduction of development cost and time”. Our ability to perform the originally planned stakeholder workshops was substantially limited by the pandemic that started in early 2020.
Bio4Comp demonstrated the scale-up of an alternative parallel computing approach that uses orders of magnitude less energy than conventional computers and thus provides an important contribution towards green computing technology. As of the time of writing of this report, to our knowledge, no further attempts to solve large combinatorial problems with DNA computing have been reported. Therefore, solving an instance of Exact Cover with 25 sets (a complexity of 33 million) constitutes a new record of the largest combinatorial problem solved with a biocomputer to date. However, quantum computing has made huge progress within the last five years. To our knowledge, using 120 logical qbits of a quantum annealer to solve Exact Cover instances with an estimated complexity of 1036, reported in April 2022, constitutes the largest combinatorial problem solved with an alternate computing approach to date.
We delivered a solid baseline for the future development of NBC by publishing a roadmap (Falco C M J M van Delft et al 2022 Nano Futures 6 032002
https://doi.org/10.1088/2399-1984/ac7d81 ), that provides benchmarks and technology targets to guide structured pre-competitive research as well as scaling models to enable prediction of how far the foreseeable and fundamentally possible improvements might take the technology.
We delivered a solid baseline for the future development of NBC by publishing a roadmap (Falco C M J M van Delft et al 2022 Nano Futures 6 032002
https://doi.org/10.1088/2399-1984/ac7d81 ), that provides benchmarks and technology targets to guide structured pre-competitive research as well as scaling models to enable prediction of how far the foreseeable and fundamentally possible improvements might take the technology.