vcrop-smoke-dataset
收藏资源简介:
该数据集包含三个配置(raw、agro_nt_1b、ntv3_650m),主要用于序列变异相关的分析或预测任务。核心数据由成对的参考序列(ref_sequence)和替代序列(alt_sequence)组成,每个样本包含一个浮点型标签(label)和一个整型折叠标识(fold)。raw配置仅提供原始序列对和标签;agro_nt_1b和ntv3_650m配置在原始序列基础上,额外提供了从特定模型(如1b、650m规模)计算得到的嵌入向量特征,包括参考序列和替代序列的均值嵌入(ref_embedding_mean, alt_embedding_mean)以及SNP位置嵌入(ref_embedding_snp, alt_embedding_snp)。ntv3_650m配置还包含余弦相似度特征(cosine_mean, cosine_snp)。所有配置的训练集均包含100个样本。数据集适用于需要序列对输入并可能利用预训练嵌入进行回归或分类的机器学习任务,例如预测序列变异的生物学效应。
This dataset includes three configurations (raw, agro_nt_1b, ntv3_650m), primarily designed for sequence variation analysis or prediction tasks. The core data consists of paired reference sequences (ref_sequence) and alternative sequences (alt_sequence), with each sample containing a floating-point label and an integer fold identifier. The raw configuration provides only the raw sequence pairs and labels; the agro_nt_1b and ntv3_650m configurations additionally offer embedding vector features computed from specific models (e.g., 1b, 650m scale), including mean embeddings for reference and alternative sequences (ref_embedding_mean, alt_embedding_mean) and SNP position embeddings (ref_embedding_snp, alt_embedding_snp). The ntv3_650m configuration also includes cosine similarity features (cosine_mean, cosine_snp). All configurations have a training set containing 100 samples. The dataset is suitable for machine learning tasks that require sequence pair inputs and may leverage pre-trained embeddings for regression or classification, such as predicting the biological effects of sequence variations.




