Predicting Adsorption Energies Using Multifidelity Data

Abstract

Composite i-PP/PEDOT films made of isotactic polypropylene (i-PP), which is frequently used for the fabrication of implantable medical devices for internal use, and chemically synthesized poly(3,4-ethylendioxythiophene) (PEDOT) nanoparticles, which are electroactive and biocompatible, have been prepared and used to detect biofilm infection. After chemical and morphological characterization, the properties (interfacial, mechanical, thermal and electrochemical) and biocompatibility of i-PP/PEDOT have been examined. Besides, carbon screen-printed electrodes coated with i-PP/PEDOT have been found to detect the growth of Gram-positive and Gram-negative bacteria through the oxidation of nicotinamide adenine dinucleotide (NADH), which comes from the bacteria metabolism (i.e. the respiration). Thus, as outer bacterial membranes are permeable to cytosolic NADH, this metabolite has been found to be an appropriated target for the detection of growing bacterial infections (biofilms). In addition, the sensor does not respond towards eukaryotic cells. This is because the major NADH pool in eukaryotic cells is located at the mitochondria and, therefore, the concentration of in the medium is not high enough to be detected since the inner mitochondrial membrane is impermeable to NADH or NAD+.

Study on the sorption mechanism of middle‐low temperature sorption thermal storage materials from the microscale simulation: A review

“…25 The data obtained from QM simulation can be divided into two types: "low fidelity" data (from DFT computations) and "high fidelity" data (from random phase approximation). 26,27 1. Classical force field The total energy in the model can be calculated by the classical force field whose parameters are determined by the empirical potential energy between different molecular states.…”

Agents for sequential learning using multiple-fidelity data

“…These models provide a clear conceptual advantage: they can combine large quantities of cheaply acquired, less accurate data with data acquired via more expensive but accurate methods. Multi-fidelity models mitigate resource limitations of building models from high fidelity data and can be significantly more accurate in their predictions than similar models trained on single-fidelity datasets 34 – 38 . For example, multi-fidelity models can combine theoretical calculations (such as DFT data) with experimental observations, or compare DFT calculations with a computationally efficient functional like PBE with a more costly but accurate functional like HSE 39 , 40 or SCAN 41 , 42 .…”

Multi-fidelity Sequential Learning for Accelerated Materials Discovery

“…While materials discovery and property prediction have been partially addressed by several machine learning strategies and frameworks [11,12,13,14,15,16,17,18,19], the diversity of materials measurement and simulation methods, along with the requirement to carry out experiments under budget constraints, require building systems that acquire new data efficiently according to their most recent results. Thus, using sequential learning (SL) and Bayesian optimization (BO) strategies [20,21,22,23,24,25,26,27] to obtain a property with experiments and/or simulations at varying fidelities and acquisition costs [28,29,30,31,32,33,34,35] are particularly promising. In SL or BO for materials design, decisions to synthesize, characterize or simulate candidate materials are made with the aid of machine learning in an iterative process.…”