alterlab-phylogenetics

系统发育学

概述

系统发育分析通过推断谱系分支模式，重建生物序列（基因、蛋白质、基因组）的进化历史。本技能涵盖标准工作流：

1. MAFFT — 多序列比对
2. IQ-TREE 3 — 通过模型选择进行最大似然树推断
3. FastTree — 快速近似最大似然（适用于大型数据集）
4. ETE3 — 用于树操作和可视化的 Python 库

安装： 比对工具和建树工具是编译型 CLI 工具（不在 PyPI 上）。通过 bioconda 或 Homebrew 安装二进制文件；使用 uv 安装 Python 可视化层。

# CLI binaries — bioconda (cross-platform) ...
conda install -c bioconda mafft iqtree fasttree trimal   # iqtree is now IQ-TREE 3
# ... or Homebrew on macOS (Apple Silicon): IQ-TREE/TrimAl live in the brewsci/bio tap
brew install mafft fasttree
brew tap brewsci/bio && brew install brewsci/bio/iqtree brewsci/bio/trimal

# Python visualization layer
uv pip install "ete3==3.1.3"   # also needs numpy<2 and PyQt5 for rendering

二进制文件名称： bioconda 的 iqtree 软件包现为 IQ-TREE 3（3.1.x），并会安装可执行文件 iqtree 和 iqtree3——其中没有 iqtree2。调用 iqtree 即可，它适用于 2.x 和 3.x 软件包；只有在 PATH 中同时存在两个版本时，才使用带版本号的名称。IQ-TREE 3 新增了 MixtureFinder、多树混合模型（MAST）、基因/位点一致性因子，以及适用于超大型病原体数据集的 CMAPLE；以下选项与 2.x 相同。

ETE3（3.1.3）是 ete3 的最后一个版本，在 Python ≥3.12 上可能难以安装（依赖旧版 numpy/PyQt5）。如果 t.render() 失败，可改为写出 Newick 树并在 FigTree/iTOL 中查看，或使用其维护中的继任者 ete4（注意：ete4 更改了 TreeStyle/render API，因此以下代码片段适用于 ete3）。

何时使用此技能

在以下情况使用系统发育分析：

• 进化关系：哪个生物/基因与我的序列亲缘关系最近？
• 病毒系统发育动力学：追踪疫情传播并估算传播日期
• 蛋白质家族分析：推断基因家族内部的进化关系
• 水平基因转移检测：识别物种树与基因树不一致的基因
• 祖先序列重建：推断祖先蛋白质序列
• 分子钟分析：利用时间采样估算分化日期
• GWAS 配套分析：将变异置于进化背景中（例如 SARS-CoV-2 变异）
• 微生物学：基于 16S rRNA 的物种系统发育或核心基因组系统发育

不适用的情况

场景	改用
操作或比较现有 Newick 树（剪枝、定根、Robinson-Foulds、重复/物种形成）	alterlab-etetoolkit
简单的序列解析、翻译或格式转换	alterlab-biopython
比对前通过同源性搜索查找序列	alterlab-blast
对群落特征表进行多样性/排序分析（UniFrac、PCoA、PERMANOVA）	alterlab-scikit-bio
运行生成 ASV 和树的 16S 扩增子工作流	alterlab-qiime2-amplicon

标准工作流

1. 使用 MAFFT 进行多序列比对

import subprocess
import os

def run_mafft(input_fasta: str, output_fasta: str, method: str = "auto",
               n_threads: int = 4) -> str:
    """
    Align sequences with MAFFT.

    Args:
        input_fasta: Path to unaligned FASTA file
        output_fasta: Path for aligned output
        method: 'auto' (auto-select), 'einsi' (accurate), 'linsi' (accurate, slow),
                'fftnsi' (medium), 'fftns' (fast), 'retree2' (fast)
        n_threads: Number of CPU threads

    Returns:
        Path to aligned FASTA file
    """
    methods = {
        "auto": ["mafft", "--auto"],
        "einsi": ["mafft", "--genafpair", "--maxiterate", "1000"],
        "linsi": ["mafft", "--localpair", "--maxiterate", "1000"],
        "fftnsi": ["mafft", "--retree", "2", "--maxiterate", "2"],
        "fftns": ["mafft", "--retree", "2", "--maxiterate", "0"],
        "retree2": ["mafft", "--retree", "2"],
    }

    cmd = methods.get(method, methods["auto"])
    cmd += ["--thread", str(n_threads), "--inputorder", input_fasta]

    with open(output_fasta, 'w') as out:
        result = subprocess.run(cmd, stdout=out, stderr=subprocess.PIPE, text=True)

    if result.returncode != 0:
        raise RuntimeError(f"MAFFT failed:\n{result.stderr}")

    # Count aligned sequences
    with open(output_fasta) as f:
        n_seqs = sum(1 for line in f if line.startswith('>'))
    print(f"MAFFT: aligned {n_seqs} sequences → {output_fasta}")

    return output_fasta

# MAFFT method selection guide:
# Few sequences (<200), accurate: linsi or einsi
# Many sequences (<1000), moderate: fftnsi
# Large datasets (>1000): fftns or auto
# Ultra-fast (>10000): mafft --retree 1

2. 修剪比对结果（可选，但建议）

def trim_alignment_trimal(aligned_fasta: str, output_fasta: str,
                            method: str = "automated1") -> str:
    """
    Trim poorly aligned columns with TrimAl.

    Methods:
    - 'automated1': Automatic heuristic (recommended)
    - 'gappyout': Remove gappy columns
    - 'strict': Strict gap threshold
    """
    cmd = ["trimal", f"-{method}", "-in", aligned_fasta, "-out", output_fasta, "-fasta"]
    result = subprocess.run(cmd, capture_output=True, text=True)
    if result.returncode != 0:
        print(f"TrimAl warning: {result.stderr}")
        # Fall back to using the untrimmed alignment
        import shutil
        shutil.copy(aligned_fasta, output_fasta)
    return output_fasta

3. IQ-TREE 3 — 最大似然树

def run_iqtree(aligned_fasta: str, output_prefix: str,
                model: str = "TEST", bootstrap: int = 1000,
                n_threads: int = 4, extra_args: list = None) -> dict:
    """
    Build a maximum likelihood tree with IQ-TREE 3.

    Args:
        aligned_fasta: Aligned FASTA file
        output_prefix: Prefix for output files
        model: 'TEST' for automatic model selection, or specify (e.g., 'GTR+G' for DNA,
               'LG+G4' for proteins, 'JTT+G' for proteins)
        bootstrap: Number of ultrafast bootstrap replicates (1000 recommended)
        n_threads: Number of threads ('AUTO' to auto-detect)
        extra_args: Additional IQ-TREE arguments

    Returns:
        Dict with paths to output files
    """
    cmd = [
        "iqtree",
        "-s", aligned_fasta,
        "--prefix", output_prefix,
        "-m", model,
        "-B", str(bootstrap),   # Ultrafast bootstrap
        "-T", str(n_threads),
        "--redo"                # Overwrite existing results
    ]

    if extra_args:
        cmd.extend(extra_args)

    result = subprocess.run(cmd, capture_output=True, text=True)

    if result.returncode != 0:
        raise RuntimeError(f"IQ-TREE failed:\n{result.stderr}")

    # Print model selection result
    log_file = f"{output_prefix}.log"
    if os.path.exists(log_file):
        with open(log_file) as f:
            for line in f:
                if "Best-fit model" in line:
                    print(f"IQ-TREE: {line.strip()}")

    output_files = {
        "tree": f"{output_prefix}.treefile",
        "log": f"{output_prefix}.log",
        "iqtree": f"{output_prefix}.iqtree",  # Full report
        "model": f"{output_prefix}.model.gz",
    }

    print(f"IQ-TREE: Tree saved to {output_files['tree']}")
    return output_files

# Model selection: ModelFinder Plus (-m MFP) is the DEFAULT — omitting -m runs it.
# -m TEST is the narrower jModelTest/ProtTest-style search (no FreeRate models);
# use it only when you need to match those tools.
# Typical winners — DNA: GTR+F+I+G4, HKY+F+G4; Protein: LG+F+G4, Q.pfam+G4;
# Codon: MG+F3X4.

# For temporal (molecular clock) analysis, add:
# extra_args = ["--date", "dates.txt", "--clock-test", "--date-CI", "95"]

4. FastTree — 快速近似 ML

适用于 IQ-TREE 速度过慢的大型数据集（>1000 条序列）：

def run_fasttree(aligned_fasta: str, output_tree: str,
                  sequence_type: str = "nt", model: str = "gtr",
                  n_threads: int = 4) -> str:
    """
    Build a fast approximate ML tree with FastTree.

    Args:
        sequence_type: 'nt' for nucleotide or 'aa' for amino acid
        model: For nt: 'gtr' (recommended) or 'jc'; for aa: 'lg', 'wag', 'jtt'
    """
    if sequence_type == "nt":
        cmd = ["FastTree", "-nt", "-gtr"]
    else:
        cmd = ["FastTree", f"-{model}"]

    cmd += [aligned_fasta]

    with open(output_tree, 'w') as out:
        result = subprocess.run(cmd, stdout=out, stderr=subprocess.PIPE, text=True)

    if result.returncode != 0:
        raise RuntimeError(f"FastTree failed:\n{result.stderr}")

    print(f"FastTree: Tree saved to {output_tree}")
    return output_tree

5. 使用 ETE3 进行树分析和可视化

from ete3 import Tree, TreeStyle, NodeStyle, TextFace, PhyloTree
import matplotlib.pyplot as plt

def load_tree(tree_file: str) -> Tree:
    """Load a Newick tree file."""
    t = Tree(tree_file)
    print(f"Tree: {len(t)} leaves, {len(list(t.traverse()))} nodes")
    return t

def basic_tree_stats(t: Tree) -> dict:
    """Compute basic tree statistics."""
    leaves = t.get_leaves()
    distances = [t.get_distance(l1, l2) for l1 in leaves[:min(50, len(leaves))]
                 for l2 in leaves[:min(50, len(leaves))] if l1 != l2]

    stats = {
        "n_leaves": len(leaves),
        "n_internal_nodes": len(t) - len(leaves),
        "total_branch_length": sum(n.dist for n in t.traverse()),
        "max_leaf_distance": max(distances) if distances else 0,
        "mean_leaf_distance": sum(distances)/len(distances) if distances else 0,
    }
    return stats

def find_mrca(t: Tree, leaf_names: list) -> Tree:
    """Find the most recent common ancestor of a set of leaves."""
    return t.get_common_ancestor(*leaf_names)

def visualize_tree(t: Tree, output_file: str = "tree.png",
                    show_branch_support: bool = True,
                    color_groups: dict = None,
                    width: int = 800) -> None:
    """
    Render phylogenetic tree to image.

    Args:
        t: ETE3 Tree object
        color_groups: Dict mapping leaf_name → color (for coloring taxa)
        show_branch_support: Show bootstrap values
    """
    ts = TreeStyle()
    ts.show_leaf_name = True
    ts.show_branch_support = show_branch_support
    ts.mode = "r"  # 'r' = rectangular, 'c' = circular

    if color_groups:
        for node in t.traverse():
            if node.is_leaf() and node.name in color_groups:
                nstyle = NodeStyle()
                nstyle["fgcolor"] = color_groups[node.name]
                nstyle["size"] = 8
                node.set_style(nstyle)

    t.render(output_file, tree_style=ts, w=width, units="px")
    print(f"Tree saved to: {output_file}")

def midpoint_root(t: Tree) -> Tree:
    """Root tree at midpoint (use when outgroup unknown)."""
    t.set_outgroup(t.get_midpoint_outgroup())
    return t

def prune_tree(t: Tree, keep_leaves: list) -> Tree:
    """Prune tree to keep only specified leaves."""
    t.prune(keep_leaves, preserve_branch_length=True)
    return t

6. 完整分析脚本

import subprocess, os
from ete3 import Tree

def full_phylogenetic_analysis(
    input_fasta: str,
    output_dir: str = "phylo_results",
    sequence_type: str = "nt",
    n_threads: int = 4,
    bootstrap: int = 1000,
    use_fasttree: bool = False
) -> dict:
    """
    Complete phylogenetic pipeline: align → trim → tree → visualize.

    Args:
        input_fasta: Unaligned FASTA
        sequence_type: 'nt' (nucleotide) or 'aa' (amino acid/protein)
        use_fasttree: Use FastTree instead of IQ-TREE (faster for large datasets)
    """
    os.makedirs(output_dir, exist_ok=True)
    prefix = os.path.join(output_dir, "phylo")

    print("=" * 50)
    print("Step 1: Multiple Sequence Alignment (MAFFT)")
    aligned = run_mafft(input_fasta, f"{prefix}_aligned.fasta",
                         method="auto", n_threads=n_threads)

    print("\nStep 2: Tree Inference")
    if use_fasttree:
        tree_file = run_fasttree(
            aligned, f"{prefix}.tree",
            sequence_type=sequence_type,
            model="gtr" if sequence_type == "nt" else "lg"
        )
    else:
        # -m TEST auto-detects the alphabet (nt vs aa) and selects the best model.
        iqtree_files = run_iqtree(
            aligned, prefix,
            model="TEST",
            bootstrap=bootstrap,
            n_threads=n_threads
        )
        tree_file = iqtree_files["tree"]

    print("\nStep 3: Tree Analysis")
    t = Tree(tree_file)
    t = midpoint_root(t)

    stats = basic_tree_stats(t)
    print(f"Tree statistics: {stats}")

    print("\nStep 4: Visualization")
    visualize_tree(t, f"{prefix}_tree.png", show_branch_support=True)

    # Save rooted tree
    rooted_tree_file = f"{prefix}_rooted.nwk"
    t.write(format=1, outfile=rooted_tree_file)

    results = {
        "aligned_fasta": aligned,
        "tree_file": tree_file,
        "rooted_tree": rooted_tree_file,
        "visualization": f"{prefix}_tree.png",
        "stats": stats
    }

    print("\n" + "=" * 50)
    print("Phylogenetic analysis complete!")
    print(f"Results in: {output_dir}/")
    return results

IQ-TREE 模型指南

DNA 模型

模型	描述	用例
GTR+G4	通用时间可逆模型 + Gamma	最灵活的 DNA 模型
HKY+G4	Hasegawa-Kishino-Yano + Gamma	双速率模型（常用）
TrN+G4	Tamura-Nei	转换率不相等
JC	Jukes-Cantor	最简单；所有速率相等

蛋白质模型

模型	描述	用例
LG+G4	Le-Gascuel + Gamma	平均表现最佳的蛋白质模型
WAG+G4	Whelan-Goldman	使用广泛
JTT+G4	Jones-Taylor-Thornton	经典模型
Q.pfam+G4	基于 Pfam 训练（QMaker）	通用蛋白质家族
Q.bird+G4	鸟类支系专用（QMaker）	鸟类蛋白质；同系列模型：Q.mammal、Q.insect、Q.yeast、Q.plant

提示： IQ-TREE 默认会自动选择模型（-m MFP）；只有在需要固定模型时才传入显式模型。

最佳实践

• 优先保证比对质量：比对质量差 → 树不可靠；请手动检查比对结果
• 小型比对（<200 条序列）使用 linsi，大型比对使用 fftns 或 auto
• 模型选择：ModelFinder Plus 是 IQ-TREE 的默认选项——省略 -m（或传入 -m MFP）即可。-m TEST 会将搜索范围限制为 jModelTest/ProtTest 风格的模型，因此仅在需要与这些工具保持一致时使用
• Bootstrap：使用 ≥1000 次 ultrafast bootstrap（-B 1000）评估分支支持度
• 为树定根：无根树可能产生误导；请使用外群定根或中点定根
• 超过 5000 条序列时使用 FastTree：IQ-TREE 会变慢；FastTree 的速度快 10–100 倍
• 修剪较长的比对：TrimAl 可移除不可靠的列，从而提高树的准确性
• 在构建树之前检查病毒/细菌序列是否发生重组（RDP4、GARD）

其他资源

• MAFFT：https://mafft.cbrc.jp/alignment/software/
• IQ-TREE 3：http://www.iqtree.org/ | 教程：https://www.iqtree.org/workshop/molevol2022
• FastTree：http://www.microbesonline.org/fasttree/
• ETE3：http://etetoolkit.org/
• FigTree（GUI 可视化）：https://tree.bio.ed.ac.uk/software/figtree/
• iTOL（网页可视化）：https://itol.embl.de/
• MUSCLE（替代比对工具）：https://www.drive5.com/muscle/
• TrimAl（比对修剪）：https://vicfero.github.io/trimal/