Xiaopeng Xu, Ph.D.

Research Scientist @ KAUST · Agentic AI for scientific discovery, protein design & synthetic biology

Recent Notes

药物设计笔记

Xiaopeng Xu published on Jun 27, 2021 included in Technology

导论

概述

创新药物发现—从基因到药物

药物 R&D

药物设计 3W

Why?
- Unmet medical needs
- Time-scale
- Success rate
What?
- Disease mechanism-Target identification & validation
- Lead generation
- Lead optimization
How?
- Drug Design –The application of CADD in drug design
- Bioinformatics, proteomics, Combi Chem/HTS
- Target-drug interaction mode
- Early ADME/T evaluation

药物设计内涵：药物研究模式的转变

Lee JA, Berg EL. J Biomol Screen. 2013 Dec;18(10):1143-55

RL 强化学习笔记

Xiaopeng Xu published on Jun 18, 2021 included in Technology

RL 基础

2 多臂赌博机 (K-arm bandit)

只有动作 (action) 和对应的收益 (rewards)。无状态 (states)。
动作价值函数
增量式实现

DL 深度学习笔记

Xiaopeng Xu published on Jun 14, 2021 included in Technology

NN 基础

二分类问题

对 64x64x3 的图片，判断是否有猫（1/0）

基本标记

X 的每一行对应的是一个样本，每一列对应的是一个特征。

逻辑回归 (Logistics regression)

SMILES 化学表示

Xiaopeng Xu published on Jun 9, 2021 included in Technology

参考：

SMILES:一种简化的分子语言：https://www.jianshu.com/p/8c915de5ad4d
https://chem.libretexts.org/Courses/University_of_Arkansas_Little_Rock/ChemInformatics_(2017)%3A_Chem_4399_5399/2.3%3A_Chemical_Representations_on_Computer%3A_Part_III

简介

SMILES，全称是 Simplified Molecular Input Line Entry System，是一种用于输入和表示分子反应的线性符号。

示例

下面看一些例子:

Metrics 评估指标

Xiaopeng Xu published on May 14, 2021 included in Technology

准确性 Accuracy

Acc. = (TP + TN)/(TP + FP + TN + FN)

精确性 Precision or Positive Predictive Value (PPV)

PPV = TP / (TP + FP)

召回率/敏感性 Sensitivity, Recall, or True Positive Rate (TPR)

TPR = TP / (TP + FN)

特异性/选择性 Specificity, Selectivity, or True Negative Rate (TNR)

TNR = TN / (TN + FP)

Negative Predictive Value (NPV)

NPV = TN / (TN + FN)

Python 常用命令

Xiaopeng Xu published on May 13, 2021 included in Technology

Pandas 列过滤

用某列对 dataframe 做过滤

papers_dates[(papers_dates.preprint_test == True) & ~ papers_dates.acptDate.isna()]
# 两列做 or 运算
df_all_norm[(df_all_norm["CO2"] > 0.8) | (df_all_norm["CO"] > 0.8)]

用日期列对 dataframe 做过滤

papers_age[papers_age.pubDate >= "2028-03-01"]

用日期间隔列对 dataframe 做过滤

papers_dates[papers_dates.preprint_age > pd.Timedelta(30,'D')]

去除空列

df.dropna(subset=['name', 'toy'])

Pandas Dataframe 合并

同长 dataframe 合并

pd.concat([papers, CSCoV_scores], axis=1)

两个 dataframe 做 left outer join

papers_redup.merge(all_rxiv_redup[['subDate', 'title']], left_on='title', right_on='title',how='left')

Pandas 数据替换

值替换

df.replace([0, 1, 2, 3], [4, 3, 2, 1])

Pandas 数据查看/分析

展示长 string

pd.set_option('display.max_colwidth', None)

查看 string 是否在某个 string list 中

X_lxd[X_lxd.title.isin(selected)][['title', 'pub_prob']]

查看日期列数据分布

papers_dates[['subDate', 'acptDate', 'pubDate']].describe(datetime_is_numeric=True)

按某列倒序排列

X_lxd.sort_values(by='pub_prob', ascending=False)

Numpy 向量计算

标量和向量的加减乘除

x = np.random.rand(3,1)
+ x # element-wise operation
- x
* x
/ x

初始化向量或者矩阵 np.zeros, np.ones

x = np.zeros((10, 1))
x = np.ones((10, 1))

Mean, std，max，min，median

x = np.random.rand(3,1)
x.mean()
x.std()
x.max()
x.min()
np.median(x)

两个向量各对应元素求最大值

x = np.random.rand(3,1)
y = np.random.rand(3,1)
np.maximum(x, y) # also np.maximum(x, 0) broadcast by default
# Out: array([[0.88240057], [0.38776741], [0.72610557]])

向量和向量的运算 *, /, dot

x = np.random.rand(3,1)
y = np.random.rand(3,1)
x * y # element-wise multiplication, return vector (3,1), or np.multiply()
np.transpose(x).dot(y) # return scalar
x / y # element-wise division, return vector (3,1)

向量绝对值 absolute

np.absolute(np.random.rand(3, 2，2)-10)

Numpy 矩阵计算

矩阵重新设置大小 reshape

x = np.random.rand(3, 2，2)
x.shape   # (3, 2，2)
x.reshape(-1, 1) # a column vector, 1 f

# reshape a np.array of images
train_set_x_flatten = train_set_x_orig.reshape(train_set_x_orig.shape[0], -1).T

scipy 相关命令

truncnorm 分布区间矫准

将标准正态分布校准到区间 [a, b] 上