HomeAIEffectively fine-tune the ESM-2 protein language mannequin with Amazon SageMaker

Effectively fine-tune the ESM-2 protein language mannequin with Amazon SageMaker

On this publish, we show easy methods to effectively fine-tune a state-of-the-art protein language mannequin (pLM) to foretell protein subcellular localization utilizing Amazon SageMaker.

Techwearclub WW

Proteins are the molecular machines of the physique, accountable for all the things from shifting your muscle tissues to responding to infections. Regardless of this selection, all proteins are manufactured from repeating chains of molecules referred to as amino acids. The human genome encodes 20 normal amino acids, every with a barely completely different chemical construction. These will be represented by letters of the alphabet, which then permits us to investigate and discover proteins as a textual content string. The big doable variety of protein sequences and buildings is what provides proteins their broad number of makes use of.

Proteins additionally play a key function in drug improvement, as potential targets but additionally as therapeutics. As proven within the following desk, lots of the top-selling medicine in 2022 had been both proteins (particularly antibodies) or different molecules like mRNA translated into proteins within the physique. Due to this, many life science researchers must reply questions on proteins quicker, cheaper, and extra precisely.

Identify Producer 2022 International Gross sales ($ billions USD) Indications
Comirnaty Pfizer/BioNTech $40.8 COVID-19
Spikevax Moderna $21.8 COVID-19
Humira AbbVie $21.6 Arthritis, Crohn’s illness, and others
Keytruda Merck $21.0 Varied cancers

Knowledge supply: Urquhart, L. High corporations and medicines by gross sales in 2022. Nature Critiques Drug Discovery 22, 260–260 (2023).

As a result of we will symbolize proteins as sequences of characters, we will analyze them utilizing strategies initially developed for written language. This contains giant language fashions (LLMs) pretrained on big datasets, which may then be tailored for particular duties, like textual content summarization or chatbots. Equally, pLMs are pre-trained on giant protein sequence databases utilizing unlabeled, self-supervised studying. We are able to adapt them to foretell issues just like the 3D construction of a protein or the way it could work together with different molecules. Researchers have even used pLMs to design novel proteins from scratch. These instruments don’t exchange human scientific experience, however they’ve the potential to hurry up pre-clinical improvement and trial design.

One problem with these fashions is their measurement. Each LLMs and pLMs have grown by orders of magnitude previously few years, as illustrated within the following determine. Which means it might take a very long time to coach them to adequate accuracy. It additionally implies that you might want to use {hardware}, particularly GPUs, with giant quantities of reminiscence to retailer the mannequin parameters.

Protein language models, like other large language models, have steadily increased in size for several years

Lengthy coaching instances, plus giant cases, equals excessive price, which may put this work out of attain for a lot of researchers. For instance, in 2023, a analysis group described coaching a 100 billion-parameter pLM on 768 A100 GPUs for 164 days! Thankfully, in lots of circumstances we will save time and sources by adapting an present pLM to our particular job. This method is named fine-tuning, and likewise permits us to borrow superior instruments from different forms of language modeling.

Resolution overview

The precise drawback we deal with on this publish is subcellular localization: Given a protein sequence, can we construct a mannequin that may predict if it lives on the skin (cell membrane) or within a cell? This is a vital piece of data that may assist us perceive the operate and whether or not it might make drug goal.

We begin by downloading a public dataset utilizing Amazon SageMaker Studio. Then we use SageMaker to fine-tune the ESM-2 protein language mannequin utilizing an environment friendly coaching methodology. Lastly, we deploy the mannequin as a real-time inference endpoint and use it to check some recognized proteins. The next diagram illustrates this workflow.

AWS architecture for fine tuning ESM

Within the following sections, we undergo the steps to arrange your coaching knowledge, create a coaching script, and run a SageMaker coaching job. All the code featured on this publish is obtainable on GitHub.

Put together the coaching knowledge

We use a part of the DeepLoc-2 dataset, which comprises a number of thousand SwissProt proteins with experimentally decided places. We filter for high-quality sequences between 100–512 amino acids:

df = pd.read_csv(
).drop(["Unnamed: 0", "Partition"], axis=1)
df["Membrane"] = df["Membrane"].astype("int32")

# filter for sequences between 100 and 512 amino acides
df = df[df["Sequence"].apply(lambda x: len(x)).between(100, 512)]

# Take away pointless options
df = df[["Sequence", "Kingdom", "Membrane"]]

Subsequent, we tokenize the sequences and break up them into coaching and analysis units:

dataset = Dataset.from_pandas(df).train_test_split(test_size=0.2, shuffle=True)
tokenizer = AutoTokenizer.from_pretrained("fb/esm2_t33_650M_UR50D")

def preprocess_data(examples, max_length=512):
    textual content = examples["Sequence"]
    encoding = tokenizer(textual content, truncation=True, max_length=max_length)
    encoding["labels"] = examples["Membrane"]
    return encoding

encoded_dataset = dataset.map(


Lastly, we add the processed coaching and analysis knowledge to Amazon Easy Storage Service (Amazon S3):

train_s3_uri = S3_PATH + "/knowledge/practice"
test_s3_uri = S3_PATH + "/knowledge/take a look at"


Create a coaching script

SageMaker script mode means that you can run your customized coaching code in optimized machine studying (ML) framework containers managed by AWS. For this instance, we adapt an present script for textual content classification from Hugging Face. This enables us to strive a number of strategies for bettering the effectivity of our coaching job.

Methodology 1: Weighted coaching class

Like many organic datasets, the DeepLoc knowledge is erratically distributed, which means there isn’t an equal variety of membrane and non-membrane proteins. We might resample our knowledge and discard data from the bulk class. Nonetheless, this would cut back the overall coaching knowledge and probably harm our accuracy. As a substitute, we calculate the category weights through the coaching job and use them to regulate the loss.

In our coaching script, we subclass the Coach class from transformers with a WeightedTrainer class that takes class weights under consideration when calculating cross-entropy loss. This helps stop bias in our mannequin:

class WeightedTrainer(Coach):
    def __init__(self, class_weights, *args, **kwargs):
        self.class_weights = class_weights
        tremendous().__init__(*args, **kwargs)

    def compute_loss(self, mannequin, inputs, return_outputs=False):
        labels = inputs.pop("labels")
        outputs = mannequin(**inputs)
        logits = outputs.get("logits")
        loss_fct = torch.nn.CrossEntropyLoss(
            weight=torch.tensor(self.class_weights, gadget=mannequin.gadget)
        loss = loss_fct(logits.view(-1, self.mannequin.config.num_labels), labels.view(-1))
        return (loss, outputs) if return_outputs else loss

Methodology 2: Gradient accumulation

Gradient accumulation is a coaching method that permits fashions to simulate coaching on bigger batch sizes. Sometimes, the batch measurement (the variety of samples used to calculate the gradient in a single coaching step) is proscribed by the GPU reminiscence capability. With gradient accumulation, the mannequin calculates gradients on smaller batches first. Then, as a substitute of updating the mannequin weights instantly, the gradients get accrued over a number of small batches. When the accrued gradients equal the goal bigger batch measurement, the optimization step is carried out to replace the mannequin. This lets fashions practice with successfully larger batches with out exceeding the GPU reminiscence restrict.

Nonetheless, additional computation is required for the smaller batch ahead and backward passes. Elevated batch sizes by way of gradient accumulation can decelerate coaching, particularly if too many accumulation steps are used. The purpose is to maximise GPU utilization however keep away from extreme slowdowns from too many additional gradient computation steps.

Methodology 3: Gradient checkpointing

Gradient checkpointing is a way that reduces the reminiscence wanted throughout coaching whereas maintaining the computational time cheap. Massive neural networks take up a number of reminiscence as a result of they should retailer all of the intermediate values from the ahead go in an effort to calculate the gradients through the backward go. This may trigger reminiscence points. One answer is to not retailer these intermediate values, however then they should be recalculated through the backward go, which takes a number of time.

Gradient checkpointing gives a balanced strategy. It saves solely a few of the intermediate values, referred to as checkpoints, and recalculates the others as wanted. Due to this fact, it makes use of much less reminiscence than storing all the things, but additionally much less computation than recalculating all the things. By strategically choosing which activations to checkpoint, gradient checkpointing permits giant neural networks to be educated with manageable reminiscence utilization and computation time. This vital method makes it possible to coach very giant fashions that might in any other case run into reminiscence limitations.

In our coaching script, we activate gradient activation and checkpointing by including the required parameters to the TrainingArguments object:

from transformers import TrainingArguments

training_args = TrainingArguments(

Methodology 4: Low-Rank Adaptation of LLMs

Massive language fashions like ESM-2 can include billions of parameters which might be costly to coach and run. Researchers developed a coaching methodology referred to as Low-Rank Adaptation (LoRA) to make fine-tuning these big fashions extra environment friendly.

The important thing thought behind LoRA is that when fine-tuning a mannequin for a particular job, you don’t must replace all the unique parameters. As a substitute, LoRA provides new smaller matrices to the mannequin that remodel the inputs and outputs. Solely these smaller matrices are up to date throughout fine-tuning, which is way quicker and makes use of much less reminiscence. The unique mannequin parameters keep frozen.

After fine-tuning with LoRA, you possibly can merge the small tailored matrices again into the unique mannequin. Or you possibly can hold them separate if you wish to rapidly fine-tune the mannequin for different duties with out forgetting earlier ones. General, LoRA permits LLMs to be effectively tailored to new duties at a fraction of the standard price.

In our coaching script, we configure LoRA utilizing the PEFT library from Hugging Face:

from peft import get_peft_model, LoraConfig, TaskType
import torch
from transformers import EsmForSequenceClassification

mannequin = EsmForSequenceClassification.from_pretrained(

peft_config = LoraConfig(

mannequin = get_peft_model(mannequin, peft_config)

Submit a SageMaker coaching job

After you’ve outlined your coaching script, you possibly can configure and submit a SageMaker coaching job. First, specify the hyperparameters:

hyperparameters = {
    "model_id": "fb/esm2_t33_650M_UR50D",
    "epochs": 1,
    "per_device_train_batch_size": 8,
    "gradient_accumulation_steps": 4,
    "use_gradient_checkpointing": True,
    "lora": True,

Subsequent, outline what metrics to seize from the coaching logs:

metric_definitions = [
    {"Name": "epoch", "Regex": "'epoch': ([0-9.]*)"},
        "Identify": "max_gpu_mem",
        "Regex": "Max GPU reminiscence use throughout coaching: ([0-9.e-]*) MB",
    {"Identify": "train_loss", "Regex": "'loss': ([0-9.e-]*)"},
        "Identify": "train_samples_per_second",
        "Regex": "'train_samples_per_second': ([0-9.e-]*)",
    {"Identify": "eval_loss", "Regex": "'eval_loss': ([0-9.e-]*)"},
    {"Identify": "eval_accuracy", "Regex": "'eval_accuracy': ([0-9.e-]*)"},

Lastly, outline a Hugging Face estimator and submit it for coaching on an ml.g5.2xlarge occasion sort. It is a cost-effective occasion sort that’s extensively obtainable in lots of AWS Areas:

from sagemaker.experiments.run import Run
from sagemaker.huggingface import HuggingFace
from sagemaker.inputs import TrainingInput

hf_estimator = HuggingFace(
    tags=[{"Key": "project", "Value": "esm-fine-tuning"}],

with Run(
) as run:
            "practice": TrainingInput(s3_data=train_s3_uri),
            "take a look at": TrainingInput(s3_data=test_s3_uri),

The next desk compares the completely different coaching strategies we mentioned and their impact on the runtime, accuracy, and GPU reminiscence necessities of our job.

Configuration Billable Time (min) Analysis Accuracy Max GPU Reminiscence Utilization (GB)
Base Mannequin 28 0.91 22.6
Base + GA 21 0.90 17.8
Base + GC 29 0.91 10.2
Base + LoRA 23 0.90 18.6

All the strategies produced fashions with excessive analysis accuracy. Utilizing LoRA and gradient activation decreased the runtime (and price) by 18% and 25%, respectively. Utilizing gradient checkpointing decreased the utmost GPU reminiscence utilization by 55%. Relying in your constraints (price, time, {hardware}), one in all these approaches could make extra sense than one other.

Every of those strategies carry out nicely by themselves, however what occurs after we use them together? The next desk summarizes the outcomes.

Configuration Billable Time (min) Analysis Accuracy Max GPU Reminiscence Utilization (GB)
All strategies 12 0.80 3.3

On this case, we see a 12% discount in accuracy. Nonetheless, we’ve lowered the runtime by 57% and GPU reminiscence use by 85%! It is a large lower that permits us to coach on a variety of cost-effective occasion sorts.

Clear up

In the event you’re following alongside in your individual AWS account, delete the any real-time inference endpoints and knowledge you created to keep away from additional fees.


bucket = boto_session.useful resource("s3").Bucket(S3_BUCKET)


On this publish, we demonstrated easy methods to effectively fine-tune protein language fashions like ESM-2 for a scientifically related job. For extra details about utilizing the Transformers and PEFT libraries to coach pLMS, try the posts Deep Studying With Proteins and ESMBind (ESMB): Low Rank Adaptation of ESM-2 for Protein Binding Website Prediction on the Hugging Face weblog. It’s also possible to discover extra examples of utilizing machine studying to foretell protein properties within the Superior Protein Evaluation on AWS GitHub repository.

In regards to the Writer

Brian Loyal Brian Loyal is a Senior AI/ML Options Architect within the International Healthcare and Life Sciences group at Amazon Net Providers. He has greater than 17 years’ expertise in biotechnology and machine studying, and is obsessed with serving to prospects remedy genomic and proteomic challenges. In his spare time, he enjoys cooking and consuming along with his family and friends.

Supply hyperlink

Opinion World [CPL] IN

latest articles

explore more