一个小型且快速的C++工具，用于计算给定fasta格式的基因序列之间的成对距离。

Python接口

要使用Python接口，您应从PyPI安装它

python -m pip install hammingdist

距离矩阵

然后，您可以从Python中以以下方式使用它

import hammingdist

# To see the different optional arguments available:
help(hammingdist.from_fasta)

# To import all sequences from a fasta file
data = hammingdist.from_fasta("example.fasta")

# To import only the first 100 sequences from a fasta file
data = hammingdist.from_fasta("example.fasta", n=100)

# To import all sequences and remove any duplicates
data = hammingdist.from_fasta("example.fasta", remove_duplicates=True)

# To import all sequences from a fasta file, also treating 'X' as a valid character
data = hammingdist.from_fasta("example.fasta", include_x=True)

# The distance data can be accessed point-wise, though looping over all distances might be quite inefficient
print(data[14,42])

输出格式

构建的距离矩阵可以写入磁盘的几种不同格式

# The data can be written to disk in csv format (default `distance` Ripser format) and retrieved:
data.dump("backup.csv")
retrieval = hammingdist.from_csv("backup.csv")

# It can also be written in lower triangular format (comma-delimited row-major, `lower-distance` Ripser format):
data.dump_lower_triangular("lt.txt")
retrieval = hammingdist.from_lower_triangular("lt.txt")

# Or in sparse format (`sparse` Ripser format: space-delimited triplet of `i j d(i,j)`
# with one line for each distance entry i > j which is not above threshold):
data.dump_sparse("sparse.txt", threshold=3)

# If the `remove_duplicates` option was used, the sequence indices can also be written.
# For each input sequence, this prints the corresponding index in the output:
data.dump_sequence_indices("indices.txt")

# The lower-triangular distance elements can also be directly accessed as a 1-d numpy array:
lt_array = data.lt_array
# The elements in this array correspond to the 2-d indices (row=1,col=0), (row=2,col=0), (row=2,col=1), ...
# These indices can be generated using the numpy tril_indices function, e.g. to construct the lower-triangular matrix:
lt_matrix = np.zeros((n_seq, n_seq))
lt_matrix[np.tril_indices(n_seq, -1)] = lt_array

重复项

当使用选项remove_duplicates=True调用from_fasta时，在构建差异矩阵之前会删除重复序列。

例如，给定以下三个输入序列

距离矩阵将是ACG和TAG之间的距离的2x2矩阵

距离矩阵的行与原始序列中的每个索引对应

最后一列是DataSet.dump_sequence_indices写入磁盘的内容。

您也可以使用hammingdist.fasta_sequence_indices而不计算距离矩阵来构建它（作为numpy数组）

import hammingdist

sequence_indices = hammingdist.fasta_sequence_indices(fasta_file)

最大距离值

默认情况下，hammingdist.from_fasta返回的距离矩阵中的元素具有最大值255。您还可以使用max_distance参数设置较小的最大值。对于大于此值的距离，hammingdist.from_fasta_large支持高达65535的距离（但需要两倍的RAM）

与参考序列的距离

可以使用以下方法计算fasta文件中每个序列与给定参考序列的距离

import hammingdist

distances = hammingdist.fasta_reference_distances(sequence, fasta_file, include_x=True)

此函数返回一个包含每个序列与参考序列距离的numpy数组。

您还可以计算两个单独序列之间的距离

import hammingdist

distance = hammingdist.distance("ACGTX", "AAGTX", include_x=True)

Linux上的OpenMP

在Linux上，hammingdist是带有OpenMP（多线程）支持的构建的，并将自动使用所有可用的CPU线程。

Linux上的CUDA

在Linux上，hammingdist还带有CUDA（Nvidia GPU）支持。要使用GPU而不是CPU，请在调用from_fasta时设置use_gpu=True。在这里，我们还设置了最大距离为2

import hammingdist

data = hammingdist.from_fasta("example.fasta", use_gpu=True, max_distance=2)

此外，现在可以直接使用GPU使用from_fasta_to_lower_triangular函数从fasta文件构建下三角矩阵文件。这避免了在内存中存储整个距离矩阵，并且将GPU上的计算与CPU上的磁盘I/O交替进行，这意味着它需要更少的RAM并运行得更快。

import hammingdist

hammingdist.from_fasta_to_lower_triangular('input_fasta.txt', 'output_lower_triangular.txt', use_gpu=True, max_distance=2)

性能历史

hammingdist中不同性能改进的影响的粗略度量