Malondialdehyde

Oxidative modifications products were assessed both in the plasma and tissue homogenates.
Advanced Oxidation Protein Products (AOPP) were estimated colorimetrically using a method Kalousová et al. (2002 (link)), which measures the total iodide ion oxidizing capacity of the samples. Absorbance at 340 nm was measured immediately by Infinite M200 PRO Multimode Microplate Reader, Tecan.
Advanced glycation end products (AGE) were estimated spectrofluorimetrically at the excitation and emission wavelengths of 350 and 440 nm using Infinite M200 PRO Multimode Microplate Reader, Tecan. Results were expressed as fluorescence/mg of the total protein.
The content of dityrosine, kynurenine, N-formylkynurenine and tryptophan was analyzed spectrofluorimetrically on 96-well microplates measuring the characteristic fluorescence at 330/415, 365/480, 325/434, and 95/340 nm respectively by Infinite M200 PRO Multimode Microplate Reader, Tecan. Results were expressed as fluorescence/mg of the total protein.
Lipid peroxidation was estimated colorimetrically using the Thiobarbituric Acid Reactive Substances (TBARS) method for measuring a malondialdehyde (MDA). 1,3,3,3 tetraethoxypropane was used as a standard (Buege and Aust, 1978 (link)).
The concentration of 4-hydroxynonenal (4-HNE) protein adducts was measured by commercial enzyme-linked immunosorbent assay (ELISA) according to the manufacturer's instructions (OxiSelect™ HNE Adduct Competitive ELISA Kit, Cell Biolabs, Inc. San Diego, CA, USA). The quantity of 4-HNE protein adducts was determined colorimetrically from a calibration curve for 4-HNE-BSA.
Total oxidant status (TOS) was measured colorimetrically based on the oxidation of ferrous ion (Fe²⁺) to ferric ion (Fe³⁺) in the presence of oxidants comprised in a sample (Erel, 2005 (link)). Changes in absorbance of the reaction solution were measured bichromatically (560/800 nm) in triplicate samples. The results were expressed as micromolar hydrogen peroxide (H₂O₂) equivalent per mg of the total protein (μmol H₂O₂ Equiv/mg of the total protein).
Oxidative stress index (OSI) was calculated according to the formula: OSI = TOS/TAC·100% (Knaś et al., 2016 (link)).
The total protein content was determined colorimetrically using the bicinchoninic acid assay (BCA assay) with bovine serum albumin (BSA) as a standard (Thermo Scientific PIERCE BCA Protein Assay Kit, Rockford, IL, USA).
All assays were performed in duplicate samples, except for the TOS determination (see above) and converted to mg of the total protein. Graphical representation of the experiment was presented on Figure 1.

The molecular energies of the various data sets are predicted using a deep tensor neural network. The core idea is to represent atoms in the molecule as vectors depending on their type and to subsequently refine the representation by embedding the atoms in their neighbourhood. This is done in a sequence of interaction passes, where the atom representations influence each other in a pair-wise fashion. While each of these refinements depends only on the pair-wise atomic distances, multiple passes enable the architecture to also take angular information into account. Because of this decomposition of atomic interactions, an efficient representation of embedded atoms is learned following quantum-chemical principles.
In the following, we describe the deep tensor neural network step-by-step, including hyper-parameters used in our experiments.
1. Assign initial atomic descriptors
We assign an initial coefficient vector to each atom i of the molecule according to its nuclear charge Z_i:

where B is the number of basis functions. All presented models use atomic descriptors with 30 coefficients. We initialize each coefficient randomly following .
2. Gaussian feature expansion of the inter-atomic distances
The inter-atomic distances D_ij are spread across many dimensions by a uniform grid of Gaussians

with Δμ being the gap between two Gaussians of width σ.
In our experiments, we set both to 0.2 Å. The centre of the first Gaussian μ_min was set to −1, while μ_max was chosen depending on the range of distances in the data (10 Å for GDB-7 and benzene, 15 Å for toluene, malonaldehyde and salicylic acid and 20 Å for GDB-9).
3. Perform T interaction passes
Each coefficient vector , corresponding to atom i after t passes, is corrected by the interactions with the other atoms of the molecule:

Here, we model the interaction v as follows:

where the circle () represents the element-wise matrix product. The factor representation in the presented models employs 60 neurons.
4. Predict energy contributions
Finally, we predict the energy contributions E_i from each atom i. Employing two fully-connected layers, for each atom a scaled energy contribution is predicted:


In our experiments, the hidden layer o_i possesses 15 neurons. To obtain the final contributions, is shifted to the mean E_μ and scaled by the s.d. E_σ of the energy per atom estimated on the training set.

This procedure ensures a good starting point for the training.
5. Obtain the molecular energy E=∑_iE_iThe bias parameters as well as are initially set to zero. All other weight matrices are initialized drawing from a uniform distribution according to (ref. 51 ). Neural network code is available.
The deep tensor neural networks have been trained for 3,000 epochs minimizing the squared error, using stochastic gradient descent with 0.9 momentum and a constant learning rate52 . The final results are taken from the models with the best validation error in early stopping.
All DTNN models were trained and executed on an NVIDIA Tesla K40 GPU. The computational cost of the employed models depends on the number of reference calculations, the number of interaction passes as well as the number of atoms per molecule. The training times for all models and data sets are shown in Supplementary Table 2, ranging from 6 h for 5.768 reference calculations of GDB-7 with one interaction pass, to 162 h for 100,000 reference calculations of the GDB-9 data set with three interaction passes.
On the other hand, the prediction is instantaneous: all models predict examples from the employed data sets in <1 ms. Supplementary Fig. 7 shows the scaling of the prediction time with the number of atoms and interaction layers. Even for a molecule with 100 atoms, a DTNN with three interaction layers requires <5 ms for a prediction.
The prediction as well as the training steps scale linearly with the number of interaction passes and quadratically with the number of atoms, since the pairwise atomic distances are required for the interactions. For large molecules it is reasonable to introduce a distance cutoff. In that case, the DTNN will also scale linearly with the number of atoms.

Thiobarbituric acid reactive substances (TBAR) level was examined to reflect the production of toxic aldehyde resulting from oxidative fatty acyl degradation, the malondialdehyde (MDA). The detailed protocol was described previously [45 (link), 46 (link)].
γ-H2AX assay of DNA damages. ARPE-19 cells were fixed in ice-cold ethanol, which were then incubated with a mouse monoclonal anti-γ-H2AX antibody (Santa Cruz Biotech). Afterwards, the FITC-conjugated anti-mouse secondary antibody (Santa Cruz) was then added. Cells were then subjected to FACS assay to determine the γ-H2AX percentage, reflecting DNA damage intensity [47 (link)].

The concentration of malondialdehyde (MDA) as a marker of oxidative stress and the levels of total antioxidant capacity (TAC), catalase (CAT) and superoxide dismutase (SOD) activity were measured using commercial kits and according to the manufacturer’s protocol (Kiazist, Iran). Briefly, kidney tissues were homogenized in lysis buffer containing protease inhibitors (Sigma–Aldrich, USA). After centrifugation by a 3-18KS Sigma centrifuge (Sigma, Germany), supernatants were collected for next analysis. MDA level was quantified by measuring thiobarbituric acid reactive substances produced in the reaction of MDA with thiobarbituric acid. TAC level was measured based on the capacity to convert Cu²⁺ to Cu⁺ ion. The activity of catalase was determined according to the reaction of the enzyme with methanol in the presence of hydrogen peroxide and measurement of generated formaldehyde. SOD activity was assayed by measuring the dismutation of superoxide radicals generated by the xanthine/xanthine oxidase system. Protein concentration of lysates was measured using Bradford method. Then the levels of oxidative stress markers were normalized to protein content [20 (link), 21 (link)].

HepG2 cells at the log phase were prepared as single-cell suspensions and seeded into 6-well plates (1 × 10⁶ cells/well) at 37°C for 24 h. After treatment with the EAF for 24 h, the cells were incubated with H₂O₂ for 4 h, washed two times with a PBS solution and lysed in lysis buffer (Biyuntian Biotechnology Co., Ltd, Shanghai, China). A BCA protein assay kit (Jiancheng Bioengineering Institute, Nanjing, China) was used to measure the intracellular protein content. The malondialdehyde (MDA) content, total antioxidant ability (T-AOC), catalase (CAT), glutathione peroxidase (GSH-Px), superoxide dismutase (SOD), Na⁺/K⁺-adenosine triphosphate (Na⁺/K⁺-ATP) and Ca²⁺/Mg²⁺-adenosine triphosphate (Ca²⁺/Mg²⁺-ATP) enzyme activity were determined with the corresponding assay kits according to the manufacturer's instructions (Jiancheng Bioengineering Institute, Nanjing, China).

A 0.30–0.45 g weighted pieces of tissue samples collected were homogenized in 50 mM phosphate buffer (pH 7.4) and centrifuged for 10 min at 3000 rpm at 4 °C and the supernatants were then stored at −20 °C until used for determination of malondialdehyde (MDA), reduced glutathione (GSH) and antioxidant enzymes CAT and SOD in accordance to following protocols described by Ohkawa et al.^{63 (link)}, Beutter et al.^{64 (link)}, Aebi^{65 (link)}, and Nishikimi et al.^{66 (link)}.