Spectral deep learning for prediction and prospective validation of functional groups

Fine, Jonathan; Rajasekar, Anand A.; Jethava, Krupal P.; Chopra, Gaurav

doi:10.1039/c9sc06240h

Cited by 85 publications

(113 citation statements)

References 49 publications

Supporting

Mentioning

110

Contrasting

Order By: Relevance

“…Recently, the combined input of FTIR and MS spectra was used to train a deep learning model to recognize the presence of functional groups such as carboxylic acid, aromaticity, and the ester group. 116 The authors concluded that FTIR spectra could in many cases reliably annotate such functional groups, but MS did offer additional information in a fair number of cases. In another recent example, infrared ion spectroscopy linked to mass spectrometry could readily separate enantiomeric N -acetylhexosamines identified in body fluid samples.…”

Section: Other Analytical Methodsmentioning

confidence: 99%

Advances in decomposing complex metabolite mixtures using substructure- and network-based computational metabolomics approaches

et al. 2021

View full text Add to dashboard Cite

show abstract

Section: Other Analytical Methodsmentioning

confidence: 99%

Advances in decomposing complex metabolite mixtures using substructure- and network-based computational metabolomics approaches

et al. 2021

View full text Add to dashboard Cite

show abstract

“…(They are not the only choices, and the research literature provides other alternatives, such as support vector machines and neural networks, that may be more appropriate vibrational spectral analysis. 12,13 ) The student handout and supporting information provide brief discussions of the assumptions of each of these models. The multiclass classification performed in Part III of this activity is structured as multiple one-vs-all binary 4 classifications, in which the probability of membership in each class is separately determined for each molecule and the class with the highest probability is then taken as the final predicted label.…”

Section: Four Common ML Classification Algorithms Are Implemented In This Exercise: Decisionmentioning

confidence: 99%

“…9 More specifically, ML has been applied to both infrared (IR) absorption and Raman vibrational spectroscopy, 10,11 including functional group identification. [12][13][14][15] Although the chemical education community acknowledges the need for student training in computational methods and ML, 16 there are limited pedagogical materials and no standard way of incorporating this into the curriculum. One approach has been the development of dedicated semester-long courses in scientific computing for chemists 17 or cheminformatics 18 that introduce programming in general and include modules on ML methods.…”

mentioning

confidence: 99%

Machine learning for functional group identification in vibrational spectroscopy: A pedagogical lab for undergraduate chemistry students

Thrall

Lee

Schrier

et al. 2021

Preprint

View full text Add to dashboard Cite

Here we report a computational activity that introduces undergraduate physical chemistry students to machine learning (ML) in the context of vibrational spectroscopy. In the first part of the activity, students use ML binary classification algorithms to distinguish between carbonyl-containing and non-carbonyl-containing molecules on the basis of their infrared absorption spectra and test modifications to this basic analysis. In a further extension of the activity, students implement a multiclass classification to predict whether carbonyl-containing molecules contain a ketone, a carboxylic acid, or another carbonyl group. This activity is designed to introduce students both to the basic workflow of a ML classification analysis and to some of the ways in which machine learning analyses can fail. We provide a comprehensive handout for the activity, including theoretical background and a detailed protocol, as well as datasets and code to implement the exercise in Python or Mathematica.

show abstract

“…An autonomous robotic experimentation platform requires both in situ and algorithmically quantifiable measurements. While efforts have gone into AI-based pattern recognition methods for advanced automated spectral analytics, [196,197] the more manageable output parameters may be derived from absorption-first excitonic peak, absorption peak line-width, concentration-and photoluminescence-peak emission energy, emission linewidth, quantum yield-spectroscopy on QDs.…”

Section: Aidriven Accelerated Materials Discovery and Optimizationmentioning

confidence: 99%

Microfluidic Synthesis of Semiconductor Materials: Toward Accelerated Materials Development in Flow

Campbell

Bateni

Volk

et al. 2020

Part & Part Syst Charact

View full text Add to dashboard Cite

Semiconductors are intriguing due to their unique electrical properties, particularly the behavior of their electrons in the presence of different stimuli (e.g., electric field, magnetic field, light irradiation), which differ greatly from conducting (i.e., metals) and insulating materials. In insulators and semiconductors, the available electronic states are discontinuous, with the existence of a gap between the lower energy states, commonly referred to as the valence band (VB), and the higher energy states, known as the conduction band (CB). The distinguishing factor between these materials is the size of the energy difference between the highest energy state in the VB and the lowest energy state in the CB, called the band gap. In insulators, the band gap is very large, making it difficult, if not impossible, to cause an electron to move from the VB to the CB. By comparison, semiconductors possess narrow to moderate band gaps, implying that it is possible to introduce enough energy to an electron to cause it to move from the VB to the CB, enabling electron transfer, the use of high energy electrons, or the emission of energy (i.e., photons) when the electron returns from the CB to the VB. The wide range of industrial application-specific requirements demands versatility in the target characteristics of semiconductor materials (e.g., size, morphology, composition, band gap, emission wavelength), which are vastly different from those commonly used in thin-film transistors. Typically, the semiconductor materials synthesized for use in the above applications are particulate materials produced across a range of sizes spanning five orders of magnitude (from just a few nm to hundreds of µm), and can be divided into two overarching categories, nano-and microsized particulate materials. Nanosized semiconductor materials, due in large part to their miniscule dimensions, present a variety of intriguing characteristics not observed in microsized semiconductor particles. First and foremost, if the dimensions of the semiconductor material decrease below a certain threshold known as the Bohr radius, the absorption and emission energies become highly dependent on the particle size, a phenomenon known as quantum confinement. [28-30] Moreover, nanosized structures have significantly larger surface-to-volume ratios which increases the surface contribution to the total free energy of the semiconductor materials, making them highly soluble [29] and therefore much easier to process and handle. Thus far, nanosized semiconductor particles have been effectively utilized in sensors, [7] LEDs, [9] Controlled synthesis of semiconductor nano/microparticles has attracted sub stantial attention for use in numerous applications from photovoltaics to photo catalysis and bioimaging due to the breadth of available physicochemical and optoelectronic properties. Microfluidic material synthesis strategies have recently been demonstrated as an effective technique for rapid development, controlled synthesis, and continuous manufacturing of solutionpr...

show abstract

Spectral deep learning for prediction and prospective validation of functional groups

Abstract: A new multi-label deep neural network architecture is used to combine Infrared and mass spectra, trained on single compounds to predict functional groups, and experimentally validated on complex mixtures.

Cited by 85 publications

References 49 publications

Advances in decomposing complex metabolite mixtures using substructure- and network-based computational metabolomics approaches

Advances in decomposing complex metabolite mixtures using substructure- and network-based computational metabolomics approaches

Machine learning for functional group identification in vibrational spectroscopy: A pedagogical lab for undergraduate chemistry students

Microfluidic Synthesis of Semiconductor Materials: Toward Accelerated Materials Development in Flow

Contact Info

Product

Resources

About