估计阅读时长: 5 分钟

基因组

为了能够实现上面所描述的这种基于EC Number的不同层级的加权嵌入,我们在原来的基因组嵌入工具上添加了一个hierarchical选项,用于支持切换为层级嵌入的操作:

Imports Microsoft.VisualBasic.Data.Framework
Imports Microsoft.VisualBasic.Data.NLP
Imports SMRUCC.genomics.Interops.NCBI.Extensions.Pipeline

Public Class GenomeMetabolicEmbedding

    ReadOnly vec As New TFIDF
    ReadOnly taxonomy As New Dictionary(Of String, String)
    ReadOnly hierarchical As Boolean = False

    Sub New(Optional hierarchical As Boolean = False)
        Me.hierarchical = hierarchical
    End Sub

    Public Sub Add(genome As GenomeVector)
        If hierarchical Then
            Call vec.Add(genome.assembly_id, genome.GetHierarchicalECNumberTerms)
        Else
            Call vec.Add(genome.assembly_id, genome.terms)
        End If

        Call taxonomy.Add(genome.assembly_id, genome.taxonomy)
    End Sub

    Public Function AddGenomes(seqs As IEnumerable(Of GenomeVector)) As GenomeMetabolicEmbedding
        For Each annotation As GenomeVector In seqs
            Call Add(annotation)
        Next

        Return Me
    End Function

    Public Function TfidfVectorizer(Optional normalize As Boolean = False) As DataFrame
        Call $"Make metabolic embedding with: ".info
        Call $"  * {vec.N} genomes".debug
        Call $"  * {vec.Words.Length} total enzyme terms".debug
        Call VBDebugger.EchoLine("")

        Dim df As DataFrame = vec.TfidfVectorizer(normalize)
        Call df.add("taxonomy", From id As String In df.rownames Select taxonomy(id))
        Return df
    End Function

    ''' <summary>
    ''' n-gram One-hot(Bag-of-n-grams)
    ''' </summary>
    ''' <returns></returns>
    Public Function OneHotVectorizer() As DataFrame
        Return vec.OneHotVectorizer
    End Function
End Class

在上面的模块中,会调用下面的函数来生成针对EC Number的层级嵌入结果:

''' <summary>
''' used for processing of the ec number terms, make the count of the
''' hierarchical ec number terms by summing the count of the specific
''' ec number terms
''' </summary>
''' <returns></returns>
Public Function GetHierarchicalECNumberTerms() As Dictionary(Of String, Integer)
    Dim hierarchical As New Dictionary(Of String, Integer)

    For Each ec_number As KeyValuePair(Of String, Integer) In terms.SafeQuery
        Dim ec As ECNumber = ECNumber.ValueParser(ec_number.Key)

        For Each ec_term As String In ec.HierarchicalECTerms
            If Not hierarchical.ContainsKey(ec_term) Then
                hierarchical(ec_term) = ec_number.Value
            Else
                hierarchical(ec_term) += ec_number.Value
            End If
        Next
    Next

    Return hierarchical
End Function

对于基于ec number来生成层级数据,我们直接使用《酶EC编号结构解析》文章末尾所展示的层级数据生成函数来实现。

基于上面的层级数据,我们仍然是将其输入TF-IDF算法模块中进行向量化嵌入

谢桂纲

No responses yet

Leave a Reply

Your email address will not be published. Required fields are marked *

博客文章
November 2026
S M T W T F S
1234567
891011121314
15161718192021
22232425262728
2930  
  1. 代谢组学那堆名词我上次查资料时也手抄过一页,QC 样本、保留时间、m/z 这些混在一张表里,来回对照着看特别费劲。你这篇把英文缩写和中文解释排在一起,省了我不少翻文献的功夫,尤其是代谢物鉴定那几段。

  2. 文中提出的"能否基于单细胞数据分析方法来分析和可视化我的基因组功能嵌入的结果矩阵呢?"让我想到一个需要先厘清的问题:把Family层级的分类当作细胞分群标签时,功能相似性与亲缘关系可能会混在一起,图上的聚集未必能直接解释为代谢功能接近。如果借用单细胞分析中的降维流程,是否可以同时标注物种分类和基因组注释完整度,看看分离主要对应哪一种差异?尤其是这里用的是EC注释的TF-IDF矩阵,缺少某项注释究竟代表功能缺失还是未被识别,也可能影响对分群结果的判断。

  3. Greetings! I stumbled upon this really useful site and figured I'd thanks for you post at here.