What We Do
Advancing the Fight Against COVID-19 through Artificial Intelligence Analysis of Chest X-Ray Images
Bringing Quantum Mechanics to Drug Discovery
Advancing quantum mechanical algorithms and applications
Advancing the Fight Against COVID-19 through Artificial Intelligence Analysis of Chest X-Ray Images.
We joined the fight against the COVID-19 pandemic by developing accurate diagnostic tools based on deep learning analysis of chest X-ray images. The technology can be deployed in the clinic and in large scale public health screenings.
We joined the fight against the COVID-19 pandemic by developing Fast-NET, FastCompChem’s deep learning algorithm to diagnose COVID-19 infection from Chest X-Ray (CXR) images. The technology can be deployed in the clinic and in large scale public health screenings.
An alternative screening method for COVID-19 infection is chest radiographic examination using CXR or Computed Tomography (CT) imaging. Early in the pandemic it was found that patients present abnormalities in chest radiography images that correlate with COVID-19 diagnosis¹ ². Some authors suggest that examination could be used as a primary tool for COVID-19 screening³.
While much of the initial use of chest imaging for COVID 19 diagnosis used CT, there are considerable disadvantages relative to CXR: 1) rapid turnout of CXR leads to more screenings per time; 2) availability of CXR equipment, including portable equipment; 3) equipment is easier to decontaminate. CXR imaging is often performed as part of standard procedure for patients with a respiratory complaint⁴.
The Fast-NET algorithm has been developed using publicly available datasets of CXRs. Initial focus was on algorithmic design and testing to establish a firm proof-of-concept. In Table 1 the performance of the algorithm on a total of 5862 CXR images is shown. Currently the algorithm has 95% accuracy in detecting COVID-19 infection with the limited data available.
Bringing Quantum Mechanics to Drug Discovery.
Quantum mechanical methods are capable of describing both nuclei and electrons, being the most accurate methodologies in chemistry and biochemistry. FastCompChem is unleashing quantum mechanics in Computer Aided Drug Design (CADD) to accelerate and lower the development costs of new drugs. FastCompChem is developing new paradigms in CADD, from virtual screening to prediction of ADME/TOX properties.
Development of new drugs is risky and costly. A recent report indicates that the overall R&D spending in the pharmaceutical/biotechnology sector has grown from around USD 128 billion in 2008 to USD 158 billion in 2017.¹
Despite these staggering numbers, the industry faces declining outcomes and immense stress. Declining R&D productivity is forcing large and small organizations to think creatively to keep up with increasing customer demand and rising costs. Relying on physical experimentation alone is not economically sustainable and in silico CADD technologies are widely used in drug discovery and development.
However, current state-of-the-art algorithms are old and based on very simplistic classical approaches, which is a serious disadvantage. Classical approximations were developed in the 1970s and 80s to allow studies of very large systems that were, and to a large extent still are, impossible to study with QM. The first modern docking study was published in 1982 by Kuntz et al.²
In the last two decades more than 60 docking/scoring programs were introduced³ but all follow the same underlying principles. In comparative studies, none has shown to be decisively superior. Use of artificial intelligence in drug discovery and development is the latest trend. Artificial intelligence algorithms, however, depend on experimental training data, that in drug discovery is scarce.⁴ ⁵
Current quantum mechanical methods have been applied sporadically to drug-design, usually in the academia.
The importance of quantum mechanics in drug discovery should not be underestimated with studies showing impactful electron transfer in solution.⁶ Since the effects are more significant on protein surfaces, it is important to include electronic effects when dealing with binding.
FastCompChem’s technologies solve major problems of quantum mechanics, including computation of electron repulsion integrals and diagonalization of very large matrices. These are cornerstones of modern quantum chemical methods. Additionally, FastCompChem also develops new and accurate semi-empirical methodologies for very large systems.
The world of molecular simulations of large systems, in particular of biological macromolecules is dominated by classical models where atoms are typically represented by charged spheres connected by springs. These are old approaches, with severe limitations, that were developed more than 40 years ago:
It is obvious that replacement of current classical methodologies with QM based technologies is long overdue. In a recent prophetic essay, Grimme and Schreiner predict low-cost QM methods will replace classical force fields in the next two decades¹.
FastCompChem developed solutions to two of the major problems that prevent extension of quantum chemical methods to very large systems, in particular computation of electron repulsion integrals² and diagonalization of very large matrices³. These are fundamental pieces that allow immediate extension of existing methodologies, for example Hartree-Fock and Density Functional Theory methods, to very large systems.
FastCompChem’s vision goes beyond the simple extension of current methodologies and aims at developing from the ground up a new semi-empirical method. Semi-empirical methods, in the broadest sense, date back from the 1960’s, with important developments in the 1980’s that are still being improved today (ex. PM7). DFTB is another example of an approximated method that is of great value today.
FastCompChem’s new algorithm, named Fast-QUANTA, starts where DFTB is short, addressing many of its limitations. It is being designed with several requirements in mind:
Besides developing new Hamiltonians, FastCompChem is introducing new parameterization strategies to the world of quantum chemistry. One of the reasons for the success of classical force field methods are the parameterization strategies. Parameterization of classical force fields, such as CHARMM, follows strict rules to ensure consistency and include reproduction of experimental condensed phase properties. This is, perhaps, the most important factor that makes inherently crude Hamiltonians usable in practice. In particular, use of condensed phase properties ensures a balanced description of the intermolecular interactions. Similarly, FastCompChem will be using the same approach.
Who We Are
The Science
We develop advanced ab initio molecular dynamics algorithms for fast simulations of large systems and new quantum-based algorithms for computer aided drug design, from virtual screening to prediction of ADME/TOX properties.
FastCompChem Lda is a software and services company that develops and provides innovative technical solutions to computer aided drug design and molecular modeling.
FastCompChem developed powerful algorithms for the accurate numerical approximation of electron repulsion integrals and linear-scaling optimization of the electronic structure that greatly simplify quantum mechanical calculations. The technology allows the development of products built on cutting-edge science to achieve high accuracy in the predictions, high performance using commonly available computer hardware while achieving great versatility in the systems that can be studied. With
FastCompChem’s algorithms is possible to study systems with 10s of thousands of atoms at ab initio and approximated levels, such as at the tight-binding level, which is an unprecedented feat. FastCompChem’s algorithms bring the precision of quantum mechanical calculations to help pharmaceutical, biotech, and academic scientists in better understanding biochemical structure and function, including new paradigms for drug discovery.
FastCompChem’s computational quantum methods follow the mantra: “one base algorithm, three core technologies and multiple applications”. The core algorithm is a tight-binding approximation that uses experimental and high-level ab initio data in the parameterization.
FastCompChem’s team is multidisciplinary with competencies in chemistry, biochemistry and computer science. Opportunistically, FastCompChem has leveraged competencies in artificial intelligence and computer graphics to develop innovative diagnostic tools to identify the Sars-COV-2 virus from readily available chest X-ray images. The technology can be used beyond clinical settings in public health applications, for example in screenings of large swaths of the population.
Pedro earned a PhD in Computational Chemistry and Reactivity from Instituto de Tecnologia Química e Biológica (ITQB-UNL), Oeiras, Portugal.
After post-doctoral work at the University of Manchester, UK, Pedro was hired by the US Centers for Disease Control and Prevention, with duty station at the University of Maryland School of Pharmacy (UMSP), where Pedro later held a faculty position at the Research Assistant professor level.
Work in Maryland consisted primarily in the development of classical force fields within CHARMM (Drude polarizable force field for proteins, CHARMM36, CgEnFF and the additive force field for silicates), and drug design projects.
Over time Pedro became increasingly aware of the need for new fast quantum mechanical approximations to replace classical force fields. Pedro developed two algorithms that tackle the major bottlenecks of computational quantum chemistry: computation of electron repulsion integrals and optimization of the electronic structure. These algorithms are foundational to FastCompChem’s projects.
Sérgio earned a PhD in Computer Science and Engineering from Beira Interior University, Covilhã, Portugal in 2015 with the thesis “Geometric Representation and Detection Methods of Protein Binding Sites”, and also has a M.Sc. in Science in Computer Engineering in 2009 (Beira Interior University).
His main research area is Bioinformatics and Computational Biology, although in a Computer Science perspective. Hence its vision is transversal to Graphic Computing and Visualization, as well as High Performance Computation through graphic processors (GPUs).
Sérgio is specialized in Visualization and Computer Graphics, Bioinformatics and Computational Biology, Parallel and GPU computing. Sérgio developed its research activities at Telecommunications Institute. Covilhã, Computer Graphics Lab (http://www.it.ubi.pt/medialab), Universidade da Beira Interior, Portugal, since June of 2006.
Vitor is a healthcare entrepreneur. Strong management skills gathered in multinationals over 18 years spent in the Industry.
Last 11 years were at Microsoft and previously at Xerox, Accenture and Bull.
He has the Advanced Management Program for Executives from Católica Business School for Business & Economics.
You can reach us from these ways:
UBIMedical, Estrada Municipal 506, 6200-284 Covilhã, Portugal
© 2021 FastCompChem