Many physiological and pathological pathways are dependent on gene-specific on/off regulation of transcription. Some genes are repressed, while others are activated. Although many previous studies have analyzed the mechanisms of gene-specific repression and activation, these studies are mainly based on the use of candidate genes, which are either repressed or activated, without simultaneously comparing and contrasting both groups of genes. There is also insufficient consideration of gene locations. Here we describe an integrated machine learning approach, using LASSO-regularized logistic regression, to model gene-specific repression and activation and the underlying contribution of chromatin interactions. LASSO-regularized logistic regression accurately predicted gene-specific transcriptional events and robustly detected the rate-limiting factors that underlie the differences of gene activation and repression. An example was provided by the leukemogenic transcription factor AML1-ETO, which is responsible for 10-15 percent of all acute myeloid leukemia cases. The analysis of AML1-ETO has also revealed novel networks of chromatin interactions and uncovered an unexpected role for E-proteins in AML1-ETO-p300 interactions and a role for the pre-existing gene state in governing the transcriptional response. Our results show that logistic regression-based probabilistic modeling is a promising tool to decipher mechanisms that integrate gene regulation and chromatin interactions in regulated transcription.

中文翻译：

---

通过 LASSO 正则化 Logistic 回归对白血病发生过程中基因特异性转录抑制、激活和染色质相互作用进行计算建模


许多生理和病理途径依赖于基因特异性的转录开/关调节。一些基因被抑制，而另一些基因被激活。尽管之前的许多研究已经分析了基因特异性抑制和激活的机制，但这些研究主要基于使用候选基因，这些基因要么被抑制，要么被激活，而没有同时比较和对比两组基因。对基因位置的考虑也不够。在这里，我们描述了一种集成的机器学习方法，使用 LASSO 正则化逻辑回归来模拟基因特异性抑制和激活以及染色质相互作用的潜在贡献。 LASSO 正则化逻辑回归准确预测了基因特异性转录事件，并稳健地检测了基因激活和抑制差异背后的限速因素。白血病转录因子 AML1-ETO 就是一个例子，它占所有急性髓系白血病病例的 10-15%。对 AML1-ETO 的分析还揭示了染色质相互作用的新网络，并揭示了 E 蛋白在 AML1-ETO-p300 相互作用中的意外作用以及预先存在的基因状态在控制转录反应中的作用。我们的结果表明，基于逻辑回归的概率模型是破译在调控转录中整合基因调控和染色质相互作用的机制的有前途的工具。