Deep Learning of CYP450 Binding of Small Molecules by Quantum Information

<p>It is currently considered by JCIM or Journal of Chemical Information and Modeling.</p>

<div><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">The data</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z s-lparen"> </span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z h-lparen">(13,443</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z"> numpy files) contains the representation of the molecular electronic properties from the curated AID1851 dataset. Out of the initial dataset, 13,443 molecules successfully underwent quantum mechanical calculations to generate the manifold embedding of molecular surfaces</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z s-lparen"> </span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z h-lparen">(MEMS),</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z"> which include electronic descriptors such as electrostatic potential</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z s-lparen"> </span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z h-lparen">(ESP),</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z"> nucleophilic and electrophilic Fukui functions</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z s-lparen"> </span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z h-lparen">(F+</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z"> and F-), and the dual descriptor</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z s-lparen"> </span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z h-lparen">(F<sup>2</sup>).</span><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z"> These quantum-based descriptors are crucial for capturing the nuanced electronic properties of the molecules, which are pivotal in understanding their interactions with CYP450 enzymes. Each file contains the MEMS for an individual molecule, effectively translating high-dimensional quantum electronic properties into a structured and lower-dimensional format, specifically designed for machine learning applications. The detailed electronic structure encoded in these numpy files allows us to predict whether a molecule is likely to inhibit specific CYP450 isozymes, such as 1A2, 2C9, 2C19, 2D6, and 3A4.</span></div>

<div> </div>

<div><span class="ace-all-bold-hthree"><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z"><b>Key features:</b></span></span></div>

<ol class="listtype-number listindent1 list-number1" start="1" style="list-style-type: decimal;">
<li><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">Each numpy file holds the MEMS derived from key electronic descriptors which are essential for capturing molecular interactions at a detailed quantum level.</span></li>
<li><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">MEMS not only encode the magnitude of these electronic properties but also their spatial distribution on the molecular surface, which is critical for understanding how molecules interact with the active sites of CYP450 enzymes.</span></li>
<li><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">The data was used to train and test deep learning models, where the numpy files served as inputs for predicting CYP450 inhibition.</span></li>
<li><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">MEMS from these files were used for shape-context analysis, enabling the comparison of the spatial distribution of electronic properties across molecules. </span></li>
</ol>

<p>该数据集目前正由JCIM（《化学信息与建模杂志》，Journal of Chemical Information and Modeling）审议。</p>
<div>该数据集（含13,443个numpy文件）包含来自经过整理的AID1851数据集的分子电子性质表征。在初始数据集中，13,443个分子成功通过量子力学计算生成了分子表面流形嵌入（MEMS，manifold embedding of molecular surfaces），其中包括静电势（ESP，electrostatic potential）、亲核与亲电Fukui函数（F+和F-，nucleophilic and electrophilic Fukui functions）以及双描述符（F²，dual descriptor）等电子描述符。这些基于量子的描述符对于捕捉分子精细的电子性质至关重要，而这些性质是理解分子与CYP450酶相互作用的关键。每个文件包含单个分子的MEMS，将高维量子电子性质有效转化为结构化的低维格式，专为机器学习应用设计。这些numpy文件中编码的详细电子结构使我们能够预测分子是否可能抑制特定的CYP450同工酶，如1A2、2C9、2C19、2D6和3A4。</div>
<div>&nbsp;</div>
<div><span class="ace-all-bold-hthree"><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z"><b>关键特征：</b></span></span></div>
<ol class="listtype-number listindent1 list-number1" start="1" style="list-style-type: decimal;">
<li><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">每个numpy文件包含从关键电子描述符衍生的MEMS，这些描述符对于在精细量子水平上捕捉分子相互作用至关重要。</span></li>
<li><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">MEMS不仅编码这些电子性质的大小，还编码其在分子表面的空间分布，这对于理解分子如何与CYP450酶的活性位点相互作用至关重要。</span></li>
<li><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">该数据用于训练和测试深度学习模型，其中numpy文件作为预测CYP450抑制作用的输入。</span></li>
<li><span class="author-d-1gg9uz65z1iz85zgdz68zmqkz84zo2qotvohlxz76zz82zz80zz80zgqq2bqz68zv4yi8dz87z10z74zhz78zxz73zz77z">这些文件中的MEMS用于形状上下文分析，能够比较不同分子间电子性质的空间分布。&nbsp;</span></li>
</ol>