Towards provably efficient quantum algorithms for large-scale machine-learning models

Large machine learning models are revolutionary technologies of artificial intelligence whose bottlenecks include huge computational expenses, power, and time used both in the pre-training and fine-tuning process. In this work, we show that fault-tolerant quantum computing could possibly provide provably efficient resolutions for generic (stochastic) gradient descent algorithms, scaling as [Formula: see text], where n is the size of the models and T is the number of iterations in the training, as long as the models are both sufficiently dissipative and sparse, with small learning rates. Based on earlier efficient quantum algorithms for dissipative differential equations, we find and prove that similar algorithms work for (stochastic) gradient descent, the primary algorithm for machine learning. In practice, we benchmark instances of large machine learning models from 7 million to 103 million parameters. We find that, in the context of sparse training, a quantum enhancement is possible at the early stage of learning after model pruning, motivating a sparse parameter download and re-upload scheme. Our work shows solidly that fault-tolerant quantum algorithms could potentially contribute to most state-of-the-art, large-scale machine-learning problems.

Analytical nonadiabatic coupling and state-specific energy gradient for the crystal field Hamiltonian describing lanthanide single-ion magnets

Paramagnetic molecules with a metal ion as an electron spin center are promising building blocks for molecular qubits and high-density memory arrays. However, fast spin relaxation and decoherence in these molecules lead to a rapid loss of magnetization and quantum information. Nonadiabatic coupling (NAC), closely related to spin-vibrational coupling, is the main source of spin relaxation and decoherence in paramagnetic molecules at higher temperatures. Predicting these couplings using numerical differentiation requires a large number of computationally intensive ab initio or crystal field electronic structure calculations. To reduce computational cost and improve accuracy, we derive and implement analytical NAC and state-specific energy gradient for the ab initio parametrized crystal field Hamiltonian describing single-ion molecular magnets. Our implementation requires only a single crystal field calculation. In addition, the accurate NACs and state-specific energy gradients can be used to model spin relaxation using sophisticated nonadiabatic molecular dynamics, which avoids the harmonic approximation for molecular vibrations. To test our implementation, we calculate the NAC values for three lanthanide complexes. The predicted values support the relaxation mechanisms reported in previous studies.

Exploring Parameter Redundancy in the Unitary Coupled-Cluster Ansätze for Hybrid Variational Quantum Computing

One of the commonly used chemically inspired approaches in variational quantum computing is the unitary coupled-cluster (UCC) ansätze. Despite being a systematic way of approaching the exact limit, the number of parameters in the standard UCC ansätze exhibits unfavorable scaling with respect to the system size, hindering its practical use on near-term quantum devices. Efforts have been taken to propose some variants of the UCC ansätze with better scaling. In this paper, we explore the parameter redundancy in the preparation of unitary coupled-cluster singles and doubles (UCCSD) ansätze employing spin-adapted formulation, small amplitude filtration, and entropy-based orbital selection approaches. Numerical results of using our approach on some small molecules have exhibited a significant cost reduction in the number of parameters to be optimized and in the time to convergence compared with conventional UCCSD-VQE simulations. We also discuss the potential application of some machine learning techniques in further exploring the parameter redundancy, providing a possible direction for future studies.

Localized Quantum Chemistry on Quantum Computers

Quantum chemistry calculations of large, strongly correlated systems are typically limited by the computation cost that scales exponentially with the size of the system. Quantum algorithms, designed specifically for quantum computers, can alleviate this, but the resources required are still too large for today's quantum devices. Here, we present a quantum algorithm that combines a localization of multireference wave functions of chemical systems with quantum phase estimation (QPE) and variational unitary coupled cluster singles and doubles (UCCSD) to compute their ground-state energy. Our algorithm, termed "local active space unitary coupled cluster" (LAS-UCC), scales linearly with the system size for certain geometries, providing a polynomial reduction in the total number of gates compared with QPE, while providing accuracy above that of the variational quantum eigensolver using the UCCSD ansatz and also above that of the classical local active space self-consistent field. The accuracy of LAS-UCC is demonstrated by dissociating (H₂)₂ into two H₂ molecules and by breaking the two double bonds in trans-butadiene, and resource estimates are provided for linear chains of up to 20 H₂ molecules.

Dissecting protein tyrosine phosphatase signaling by engineered chemogenetic control of its activity

Protein tyrosine phosphatases (PTPases) are critical mediators of dynamic cell signaling. A tool capable of identifying transient signaling events downstream of PTPases is essential to understand phosphatase function on a physiological time scale. We report a broadly applicable protein engineering method for allosteric regulation of PTPases. This method enables dissection of transient events and reconstruction of individual signaling pathways. Implementation of this approach for Shp2 phosphatase revealed parallel MAPK and ROCK II dependent pathways downstream of Shp2, mediating transient cell spreading and migration. Furthermore, we show that the N-SH2 domain of Shp2 regulates MAPK-independent, ROCK II-dependent cell migration. Engineered targeting of Shp2 activity to different protein complexes revealed that Shp2-FAK signaling induces cell spreading whereas Shp2-Gab1 or Shp2-Gab2 mediates cell migration. We identified specific transient morphodynamic processes induced by Shp2 and determined the role of individual signaling pathways downstream of Shp2 in regulating these events. Broad application of this approach is demonstrated by regulating PTP1B and PTP-PEST phosphatases.

Structure- and Interaction-Based Design of Anti-SARS-CoV-2 Aptamers

Aptamer selection against novel infections is a complicated and time-consuming approach. Synergy can be achieved by using computational methods together with experimental procedures. This study aims to develop a reliable methodology for a rational aptamer in silico et vitro design. The new approach combines multiple steps: (1) Molecular design, based on screening in a DNA aptamer library and directed mutagenesis to fit the protein tertiary structure; (2) 3D molecular modeling of the target; (3) Molecular docking of an aptamer with the protein; (4) Molecular dynamics (MD) simulations of the complexes; (5) Quantum-mechanical (QM) evaluation of the interactions between aptamer and target with further analysis; (6) Experimental verification at each cycle for structure and binding affinity by using small-angle X-ray scattering, cytometry, and fluorescence polarization. By using a new iterative design procedure, structure- and interaction-based drug design (SIBDD), a highly specific aptamer to the receptor-binding domain of the SARS-CoV-2 spike protein, was developed and validated. The SIBDD approach enhances speed of the high-affinity aptamers development from scratch, using a target protein structure. The method could be used to improve existing aptamers for stronger binding. This approach brings to an advanced level the development of novel affinity probes, functional nucleic acids. It offers a blueprint for the straightforward design of targeting molecules for new pathogen agents and emerging variants.

The role of SAXS and molecular simulations in 3D structure elucidation of a DNA aptamer against lung cancer

Aptamers are short, single-stranded DNA or RNA oligonucleotide molecules that function as synthetic analogs of antibodies and bind to a target molecule with high specificity. Aptamer affinity entirely depends on its tertiary structure and charge distribution. Therefore, length and structure optimization are essential for increasing aptamer specificity and affinity. Here, we present a general optimization procedure for finding the most populated atomistic structures of DNA aptamers. Based on the existed aptamer LC-18 for lung adenocarcinoma, a new truncated LC-18 (LC-18t) aptamer LC-18t was developed. A three-dimensional (3D) shape of LC-18t was reported based on small-angle X-ray scattering (SAXS) experiments and molecular modeling by fragment molecular orbital or molecular dynamic methods. Molecular simulations revealed an ensemble of possible aptamer conformations in solution that were in close agreement with measured SAXS data. The aptamer LC-18t had stronger binding to cancerous cells in lung tumor tissues and shared the binding site with the original larger aptamer. The suggested approach reveals 3D shapes of aptamers and helps in designing better affinity probes.

Ab initio molecular dynamics on quantum computers

Ab initio molecular dynamics (AIMD) is a valuable technique for studying molecules and materials at finite temperatures where the nuclei evolve on potential energy surfaces obtained from accurate electronic structure calculations. In this work, we present an approach to running AIMD simulations on noisy intermediate-scale quantum (NISQ)-era quantum computers. The electronic energies are calculated on a quantum computer using the variational quantum eigensolver (VQE) method. Algorithms for computation of analytical gradients entirely on a quantum computer require quantum fault-tolerant hardware, which is beyond NISQ-era. Therefore, we compute the energy gradients numerically using finite differences, the Hellmann-Feynman theorem, and a correlated sampling technique. This method only requires additional classical calculations of electron integrals for each degree of freedom without any additional computations on a quantum computer beyond the initial VQE run. As a proof of concept, AIMD simulations are demonstrated for the H₂ molecule on IBM quantum devices. In addition, we demonstrate the validity of the method for larger molecules using full configuration interaction wave functions. As quantum hardware and noise mitigation techniques continue to improve, the method can be utilized for studying larger molecular systems.

NWChem: Past, present, and future

Specialized computational chemistry packages have permanently reshaped the landscape of chemical and materials science by providing tools to support and guide experimental efforts and for the prediction of atomistic and electronic properties. In this regard, electronic structure packages have played a special role by using first-principle-driven methodologies to model complex chemical and materials processes. Over the past few decades, the rapid development of computing technologies and the tremendous increase in computational power have offered a unique chance to study complex transformations using sophisticated and predictive many-body techniques that describe correlated behavior of electrons in molecular and condensed phase systems at different levels of theory. In enabling these simulations, novel parallel algorithms have been able to take advantage of computational resources to address the polynomial scaling of electronic structure methods. In this paper, we briefly review the NWChem computational chemistry suite, including its history, design principles, parallel tools, current capabilities, outreach, and outlook.

Real-time data analysis for medical diagnosis using FPGA-accelerated neural networks

Background

Real-time analysis of patient data during medical procedures can provide vital diagnostic feedback that significantly improves chances of success. With sensors becoming increasingly fast, frameworks such as Deep Neural Networks are required to perform calculations within the strict timing constraints for real-time operation. However, traditional computing platforms responsible for running these algorithms incur a large overhead due to communication protocols, memory accesses, and static (often generic) architectures. In this work, we implement a low-latency Multi-Layer Perceptron (MLP) processor using Field Programmable Gate Arrays (FPGAs). Unlike CPUs and Graphics Processing Units (GPUs), our FPGA-based design can directly interface sensors, storage devices, display devices and even actuators, thus reducing the delays of data movement between ports and compute pipelines. Moreover, the compute pipelines themselves are tailored specifically to the application, improving resource utilization and reducing idle cycles. We demonstrate the effectiveness of our approach using mass-spectrometry data sets for real-time cancer detection.

Results

We demonstrate that correct parameter sizing, based on the application, can reduce latency by 20% on average. Furthermore, we show that in an application with tightly coupled data-path and latency constraints, having a large amount of computing resources can actually reduce performance. Using mass-spectrometry benchmarks, we show that our proposed FPGA design outperforms both CPU and GPU implementations, with an average speedup of 144x and 21x, respectively.

Conclusion

In our work, we demonstrate the importance of application-specific optimizations in order to minimize latency and maximize resource utilization for MLP inference. By directly interfacing and processing sensor data with ultra-low latency, FPGAs can perform real-time analysis during procedures and provide diagnostic feedback that can be critical to achieving higher percentages of successful patient outcomes.

Ligand binding to an Allergenic Lipid Transfer Protein Enhances Conformational Flexibility resulting in an Increase in Susceptibility to Gastroduodenal Proteolysis

Non-specific lipid transfer proteins (LTPs) are a family of lipid-binding molecules that are widely distributed across flowering plant species, many of which have been identified as allergens. They are highly resistant to simulated gastroduodenal proteolysis, a property that may play a role in determining their allergenicity and it has been suggested that lipid binding may further increase stability to proteolysis. It is demonstrated that LTPs from wheat and peach bind a range of lipids in a variety of conditions, including those found in the gastroduodenal tract. Both LTPs are initially cleaved during gastroduodenal proteolysis at three major sites between residues 39–40, 56–57 and 79–80, with wheat LTP being more resistant to cleavage than its peach ortholog. The susceptibility of wheat LTP to proteolyic cleavage increases significantly upon lipid binding. This enhanced digestibility is likely to be due to the displacement of Tyr79 and surrounding residues from the internal hydrophobic cavity upon ligand binding to the solvent exposed exterior of the LTP, facilitating proteolysis. Such knowledge contributes to our understanding as to how resistance to digestion can be used in allergenicity risk assessment of novel food proteins, including GMOs.

Potent inhibition of VEGFR-2 activation by tight binding of green tea epigallocatechin gallate and apple procyanidins to VEGF: relevance to angiogenesis

Scope

Excessive concentrations of vascular endothelial growth factor (VEGF) drive angiogenesis and cause complications such as increased growth of tumours and atherosclerotic plaques. The aim of this study was to determine the molecular mechanism underlying the potent inhibition of VEGF signalling by polyphenols.

Methods and results

We show that the polyphenols epigallocatechin gallate from green tea and procyanidin oligomers from apples potently inhibit VEGF-induced VEGF receptor-2 (VEGFR-2) signalling in human umbilical vein endothelial cells by directly interacting with VEGF. The polyphenol-induced inhibition of VEGF-induced VEGFR-2 activation occurred at nanomolar polyphenol concentrations and followed bi-phasic inhibition kinetics. VEGF activity could not be recovered by dialysing VEGF-polyphenol complexes. Exposure of VEGF to epigallocatechin gallate or procyanidin oligomers strongly inhibited subsequent binding of VEGF to human umbilical vein endothelial cells expressing VEGFR-2. Remarkably, even though VEGFR-2 signalling was completely inhibited at 1 μM concentrations of polyphenols, endothelial nitric oxide synthase was shown to still be activated via the PI3K/Akt signalling pathway which is downstream of VEGFR-2.

Conclusion

These data demonstrate for the first time that VEGF is a key molecular target for specific polyphenols found in tea, apples and cocoa which potently inhibit VEGF signalling and angiogenesis at physiological concentrations. These data provide a plausible mechanism which links bioactive compounds in food with their beneficial effects.

The structural characteristics of nonspecific lipid transfer proteins explain their resistance to gastroduodenal proteolysis

The structure and stability of the allergenic nonspecific lipid transfer protein (LTP) of peach were compared with the homologous LTP1 of barley and its liganded form LTP1b. All three proteins were resistant to gastric pepsinolysis and were only slowly digested at 1 to 2 out of 14 potential tryptic and chymotryptic cleavage sites under duodenal conditions. Peach LTP was initially cleaved at Tyr79-Lys80 and then at Arg39-Thr40 (a site lost in barley LTP1). Molecular dynamics simulations of the proteins under folded conditions showed that the backbone flexibility is limited, explaining the resistance to duodenal proteolysis. Arg39 and Lys80 side chains were more flexible in simulations of peach compared with barley LTP1. This may explain differences in the rates of cleavage observed experimentally for the two proteins and suggests that the flexibility of individual amino acid side chains could be important in determining preferred proteolytic cleavage sites. In order to understand resistance to pepsinolysis, proteins were characterized by NMR spectroscopy at pH 1.8. This showed that the helical regions of both proteins remain folded at this pH. NMR hydrogen exchange studies confirmed the rigidity of the structures at acidic pH, with barley LTP1 showing some regions with greater protection. Collectively, these data suggest that the rigidity of the LTP scaffold is responsible for their resistance to proteolysis. Gastroduodenal digestion conditions do not disrupt the 3D structure of peach LTP, explaining why LTPs retain their ability to bind IgE after digestion and hence their allergenic potential.

A parallel distributed data CPHF algorithm for analytic Hessians

One of the most commonly used means to characterize potential energy surfaces of reactions and chemical systems is the Hessian calculation, whose analytic evaluation is computationally and memory demanding. A new scalable distributed data analytic Hessian algorithm is presented. Features of the distributed data parallel coupled perturbed Hartree-Fock (CPHF) are (a) columns of density-like and Fock-like matrices are distributed among processors, (b) an efficient static load balancing scheme achieves good work load distribution among the processors, (c) network communication time is minimized, and (d) numerous performance improvements in analytic Hessian steps are made. As a result, the new code has good performance which is demonstrated on large biological systems.

Ab initio and analytic intermolecular potentials for Ar-CF4

Ab initio calculations at the CCSD(T) level of theory were performed to characterize the Ar + CF4 intermolecular potential. Potential energy curves were calculated with the aug-cc-pVTZ basis set, and with and without a correction for basis set superposition error (BSSE). Additional calculations were performed with other correlation consistent basis sets to extrapolate the Ar-CF4 potential energy minimum to the complete basis set (CBS) limit. Both the size of the basis set and BSSE have substantial effects on the Ar + CF4 potential. Calculations with the aug-cc-pVTZ basis set, and with a BSSE correction, appear to give a good representation of the BSSE corrected potential at the CBS limit. In addition, MP2 theory is found to give potential energies in very good agreement with those determined by the much higher level CCSD(T) theory. Two model analytic potential energy functions were determined for Ar + CF4. One is a fit to the aug-cc-pVTZ calculations with a BSSE correction. The second was derived by fitting an average BSSE corrected potential, which is an average of the CCSD(T)/aug-cc-pVTZ potentials with and without a BSSE correction. These analytic functions are written as a sum of two-body potentials and excellent fits to the ab initio potentials are obtained by representing each two-body interaction as a Buckingham potential.

Ab initio and analytic intermolecular potentials for Ar–CH3OH

Ab initio calculations at the CCSD(T)/aug-cc-pVTZ level of theory were used to characterize the Ar-CH(3)OH intermolecular potential energy surface (PES). Potential energy curves were calculated for four different Ar + CH(3)OH orientations and used to derive an analytic function for the intermolecular PES. A sum of Ar-C, Ar-O, Ar-H(C), and Ar-H(O) two-body potentials gives an excellent fit to these potential energy curves up to 100 kcal mol(-1), and adding an additional r(-n) term to the Buckingham two-body potential results in only a minor improvement in the fit. Three Ar-CH(3)OH van der Waals minima were found from the CCSD(T)/aug-cc-pVTZ//MP2/aug-cc-pVTZ calculations. The structure of the global minimum is in overall good agreement with experiment (X.-C. Tan, L. Sun and R. L. Kuczkowski, J. Mol. Spectrosc., 1995, 171, 248). It is T-shaped with the hydroxyl H-atom syn with respect to Ar. Extrapolated to the complete basis set (CBS) limit, the global minimum has a well depth of 0.72 kcal mol(-1) with basis set superposition error (BSSE) correction. The aug-cc-pVTZ basis set gives a well depth only 0.10 kcal mol(-1) smaller than this value. The well depths of the other two minima are within 0.16 kcal mol(-1) of the global minimum. The analytic Ar-CH(3)OH intermolecular potential also identifies these three minima as the only van der Waals minima and the structures predicted by the analytic potential are similar to the ab initio structures. The analytic potential identifies the same global minimum and the predicted well depths for the minima are within 0.05 kcal mol(-1) of the ab initio values. Combining this Ar-CH(3)OH intermolecular potential with a potential for a OH-terminated alkylthiolate self-assembled monolayer surface (i.e., HO-SAM) provides a potential to model Ar + HO-SAM collisions.

Thermochemical Properties of HxNO Molecules and Ions from ab Initio Electronic Structure Theory

Coupled-cluster calculations through noniterative triple excitations were used to compute optimized structures, atomization energies at 0 K, and heats of formation at 0 and 298 K for NH2O, HNOH, NH2O-, NH2OH+, NH3OH+, HNO-, and HON. These molecules are important in the gas-phase oxidation of NH3, as well as its solution-phase chemistry. The O-H, N-H, and N-O bond energies of these molecules are given and compared. The N-H and O-H bond energies are quite low, and, for NH2OH, the O-H bond is weaker than the N-H bond (by 7.5 kcal/mol). The energetics for a variety of ionic chemical processes in the gas phase, including the electron affinities of NH2O and HNO, the proton affinities of NH2O and NH2OH, and the acidities of NH2OH and NH2O, are given. The compounds are weak bases and weak acids in the gas phase. Solvation effects were included at the PCM and COSMO levels. The COSMO model gave better values than the PCM model. The relative values for pKa for NH2O and NH2OH are in good agreement with the experimental values, showing both compounds to be very strong bases in aqueous solution with NH2OH being the stronger base by 1.8 pK units at the COSMO level, compared to the experimental pK difference of 1.1+/-0.3 pK units. We predict that NH2OH+ will not be formed in aqueous solution, because it is a very strong acid. Based on the known acidity of NH3OH+, we predict pKa(NH2OH+)=-5.4 at the COSMO level, which is in good agreement with the experimental estimate of pKa(NH2OH+)=-7+/-2.

Effect of the Ar–Ni(s) potential on the cross section for Ar+CH4/Ni{111} collision-induced desorption and the need for a more accurate CH4/Ni{111} potential

In a previous paper [L. Sun, P. de Sainte Claire, O. Meroueh, and W. L Hase, J. Chem. Phys. 114, 535 (2001)], a classical trajectory simulation was reported of CH(4) desorption from Ni{111} by Ar-atom collisions. At an incident angle theta(i) of 60 degrees (with respect to the surface normal), the calculated collision-induced desorption (CID) cross sections are in excellent agreement with experiment. However, for smaller incident angles the calculated cross sections are larger than the experimental values and for normal collisions, theta(i)=0 degrees , the calculated cross sections are approximately a factor of 2 larger. This trajectory study used an analytic function for the Ar+Ni(s) intermolecular potential which gives an Ar-Ni{111} potential energy minimum which is an order of magnitude too deep. In the work reported here, the previous trajectory study is repeated with an Ar+Ni(s) analytic intermolecular potential which gives an accurate Ar-Ni{111} potential energy minimum and also has a different surface corrugation than the previous potential. Though there are significant differences between the two Ar+Ni(s) analytic potentials, they have no important effects on the CID dynamics and the cross sections reported here are nearly identical to the previous values. Zero-point energy motions of the surface and the CH(4)-Ni(s) intermolecular modes are considered in the simulation and they are found to have a negligible effect on the CID cross sections. Calculations of the intermolecular potential between CH(4) and a Ni atom, at various levels of theory, suggest that there are substantial approximations in the ab initio calculation used to develop the CH(4)+Ni{111} potential. The implication is that the differences between the trajectory and experimental CID cross sections may arise from an inaccurate CH(4)+Ni{111} potential used in the trajectory simulation.

Component-based integration of chemistry and optimization software

Typical scientific software designs make rigid assumptions regarding programming language and data structures, frustrating software interoperability and scientific collaboration. Component-based software engineering is an emerging approach to managing the increasing complexity of scientific software. Component technology facilitates code interoperability and reuse. Through the adoption of methodology and tools developed by the Common Component Architecture Forum, we have developed a component architecture for molecular structure optimization. Using the NWChem and Massively Parallel Quantum Chemistry packages, we have produced chemistry components that provide capacity for energy and energy derivative evaluation. We have constructed geometry optimization applications by integrating the Toolkit for Advanced Optimization, Portable Extensible Toolkit for Scientific Computation, and Global Arrays packages, which provide optimization and linear algebra capabilities. We present a brief overview of the component development process and a description of abstract interfaces for chemical optimizations. The components conforming to these abstract interfaces allow the construction of applications using different chemistry and mathematics packages interchangeably. Initial numerical results for the component software demonstrate good performance, and highlight potential research enabled by this platform.