ChatGPT与DeepSeek科研对决：谁才是真正的学术加速器？

文献理解与知识汇总能力

• 从文献中任意找一段文字进行输入，对比两个模型输出结果的异同

A total of 32 CpG sites in eight different genes that had previously been associated with age were considered (Table 1). These sites are frequently incorporated in the predictive models used to estimate age from DNA samples derived from blood based on multiple forensic studies: Bekaert et al. [18]; Dias et al. [6]; Feng et al. [19]; Fleckhaus and Schneider [20]; Freire-Aradas et al. [33]; Freire-Aradas et al. [7]; Jung et al. [28]; Lee et al. [9]; Naue et al. [30]; Naue et al. [31]; Wo z ́niak et al. [12]; Zbie ́c-Piekarska, Sp ́ olnicka, Kupiec, Makowska, et al. [25]; Zbiec ́-Piekarska, Sp ́ olnicka, Kupiec, Parys-Proszek, et al. [26]. These sites, which we included in our adaptive sampling panel, were used for the assessment of nanopore methylation in our ten samples spanning ages from 25 to 76 (Supp. Table 1). To assess the relationship between the chronological age of our sample donors and methylation levels at these CpG sites, we created a linear regression model commonly used in other forensic methylation studies [18,7,9,31,12,25]. A varying degree of correlation was observed, with a majority of CpG sites showing a positive correlation with age and having r-value greater than 0.5, indicating that methylation levels at these sites generally increase with age. The site with the highestpositive correlation with age was locatedin the gene ELOVL2 (at chr6:11044644 (GRCh38)), with an r-value of 0.948 (Table 1). While several other sites showed strong positive correlations (r ≥ 0.8), a few sites showed weaker correlations with varying values of RMSE (Table 1).

• 输出

分别用到的模型为deepseek-r1和o3-mini，整体比较两者都针对给出的文献部分进行了正确的总结，但是DeepSeek相较于OpenAI的模型给出了更加详细的结果。

 

思维延伸能力

提问内容：基于以上的部分文献内容，我可以在法医有关甲基化研究课题上有什么样的延伸和创新

 

• 输出

数据处理能力

我现在有一批纳米孔测序平台产生的甲基化样本的原始下机序列数据，我要怎么进行后续分析，给出处理步骤以及代码示例

 

• 输出

总结

1. 不同的模型具有不同的优势，本文的示例仅简单展示了未进行prompt编写下的返回结果。

2. 大语言模型可以给我们科学研究带来便利，但请辩证使用以及辨别返回结果的真伪。