Performance of chemical structure string representations for chemical image recognition using transformers

The use of molecular string representations for deep learning in chemistry has been steadily increasing in recent years. The complexity of existing string representations, and the difficulty in creating meaningful...

Quantitative Analysis of Multiple Food Additives in Wheat Flour by Raman Hyperspectral Imaging

Raman hyperspectral imaging can obtain both the internal Raman signals and the external image information of the sample simultaneously. This study investigated the quantitatively analysis of multiple food additives in wheat flour by using this technology. Raman hyperspectral images of wheat flour containing the three additives, L-ascorbate acid (LAA), azodicarbonamide (ADC) and benzoyl peroxide (BPO), were collected. Raman signals in Raman hyperspectral images were preprocessed by smoothing and baseline correction methods to obtain the corrected image. Chemical images were created to visually identify additive pixels by selecting single-band image corresponding to Raman characteristic peaks of each additive from the corrected image and combining with the threshold segmentation method. The results showed that the chemical image can identify the above three additives in wheat flour. The identified additive pixels have a significant linear relationship with their concentration, and the coefficients of determination of LAA, ADC and BPO in the quantitative model were 0.9858, 0.9868 and 0.9830, respectively. This study indicated that the Raman characteristic peaks and threshold segmentation provide a non-destructive method for quantitative analysis of multiple wheat flour additives in Raman hyperspectral images.

Performance of chemical structure string representations for chemical image recognition using transformers

The use of molecular string representations for deep learning in chemistry has been steadily increasing in recent years. The complexity of existing string representations, and the difficulty in creating meaningful tokens from them, lead to the development of new string representations for chemical structures. In this study, the translation of chemical structure depictions in the form of bitmap images to corresponding molecular string representations was examined. An analysis of the recently developed DeepSMILES and SELFIES representations in comparison with the most commonly used SMILES representation is presented where the ability to translate image features into string representations with transformer models was specifically tested. The SMILES representation exhibits the best overall performance whereas SELFIES guarantee valid chemical structures. DeepSMILES performs in between SMILES and SELFIES, InChIs are not appropriate for the learning task. All investigations were carried out with publicly available datasets and the code used to train and evaluate the models has been made available to the public.

DECIMER 1.0: deep learning for chemical image recognition using transformers

The amount of data available on chemical structures and their properties has increased steadily over the past decades. In particular, articles published before the mid-1990 are available only in printed or scanned form. The extraction and storage of data from those articles in a publicly accessible database are desirable, but doing this manually is a slow and error-prone process. In order to extract chemical structure depictions and convert them into a computer-readable format, Optical Chemical Structure Recognition (OCSR) tools were developed where the best performing OCSR tools are mostly rule-based. The DECIMER (Deep lEarning for Chemical ImagE Recognition) project was launched to address the OCSR problem with the latest computational intelligence methods to provide an automated open-source software solution. Various current deep learning approaches were explored to seek a best-fitting solution to the problem. In a preliminary communication, we outlined the prospect of being able to predict SMILES encodings of chemical structure depictions with about 90% accuracy using a dataset of 50–100 million molecules. In this article, the new DECIMER model is presented, a transformer-based network, which can predict SMILES with above 96% accuracy from depictions of chemical structures without stereochemical information and above 89% accuracy for depictions with stereochemical information.

Evaluation of Wood Quality Traits in Salix viminalis Useful for Biofuels: Characterization and Method Development

Salix (willow) is a well-known coppice plant that has been used as a source for bioenergy for decades. With recent developments in changing from a fossil-based to a circular bioeconomy, greater interest has been orientated towards willow as a potential source of biomass for transport biofuels. This has created increasing interest for breeding strategies to produce interesting genotypic and phenotypic traits in different willow varieties. In the present study, 326 genetically distinct clones and several commercial varieties of S. viminalis were analyzed using complementary approaches including density, chemical, image, histochemical, and morphometric analyses. A systematic approach was adopted whereby the basal regions of harvested stems were separated and used in all studies to aid comparisons. Density analyses were performed on all clone individuals, and from the results, 20 individual plants representing 19 clones were selected for the more in-depth analyses (chemical, image analysis, histochemical, and morphometric). The absolute dry density of the clones selected varied between ca. 300 and 660 kg/m3 with less variation seen in the commercial S. viminalis varieties (ca. 450–520 kg/m3). Selected clones for chemical analysis showed the largest variation in glucose (47.3–60.1%; i.e., cellulose) and total sugar content, which ranged between ca. 61 and 77% and only ca. 16 and 22% for lignin. Image analyses of entire basal stem sections showed presence of tension wood in variable amounts (ca. 7–39%) with characteristic G-fibers containing cellulose-rich and non-lignified gelatinous layers. Several of the clones showing prominent tension wood also showed high glucose and total sugar content as well as low lignin levels. A morphometric approach using an optical fiber analyzer (OFA) for analyzing 1000 s (minimum 100,000 particles) of macerated fibers was evaluated as a convenient tool for determining the presence of tension wood in stem samples. Statistical analyses showed that for S. viminalis stems of the same density and thickness, the OFA approach could separate tension wood fibers from normal wood fibers by length but not fiber width. Results emphasized considerable variability between the clones in the physical and chemical approaches adopted, but that a common aspect for all clones was the occurrence of tension wood. Since tension wood with G-fibers and cellulose-rich G-layers represents an increased source of readily available non-recalcitrant cellulose for biofuels, S. viminalis breeding programs should be orientated towards determining factors for its enhancement.

DECIMER 1.0: Deep Learning for Chemical Image Recognition using Transformers

<p>The amount of data available on chemical structures and their properties has increased exponentially over the past decades. In particular, articles published before the mid-1990 are available only in printed or scanned form. The extraction and storage of data from those articles in a publicly accessible database are desirable, but doing this manually is a slow and error-prone process. In order to extract chemical structure depictions and convert them into a computer-readable format, optical chemical structure recognition (OCSR) tools were developed where the best performing OCSR tools are mostly rule-based.</p><p> </p><p>The DECIMER (Deep lEarning for Chemical ImagE Recognition) project was launched to address the OCSR problem with the latest computational intelligence methods to provide an automated open-source software solution. Various current deep learning approaches were explored to seek a best-fitting solution to the problem. In a preliminary communication, we outlined the prospect of being able to predict SMILES encodings of chemical structure depictions with about 90% accuracy using a dataset of 50-100 million molecules. In this article, the new DECIMER model is presented, a transformer-based network, which can predict SMILES with above 96% accuracy from depictions of chemical structures without stereochemical information and above 89% accuracy for depictions with stereochemical information.</p><p><br></p>

DECIMER 1.0: Deep Learning for Chemical Image Recognition using Transformers

<p>The amount of data available on chemical structures and their properties has increased exponentially over the past decades. In particular, articles published before the mid-1990 are available only in printed or scanned form. The extraction and storage of data from those articles in a publicly accessible database are desirable, but doing this manually is a slow and error-prone process. In order to extract chemical structure depictions and convert them into a computer-readable format, optical chemical structure recognition (OCSR) tools were developed where the best performing OCSR tools are mostly rule-based.</p><p> </p><p>The DECIMER (Deep lEarning for Chemical ImagE Recognition) project was launched to address the OCSR problem with the latest computational intelligence methods to provide an automated open-source software solution. Various current deep learning approaches were explored to seek a best-fitting solution to the problem. In a preliminary communication, we outlined the prospect of being able to predict SMILES encodings of chemical structure depictions with about 90% accuracy using a dataset of 50-100 million molecules. In this article, the new DECIMER model is presented, a transformer-based network, which can predict SMILES with above 96% accuracy from depictions of chemical structures without stereochemical information and above 89% accuracy for depictions with stereochemical information.</p><p><br></p>